\documentclass[11pt]{article} \usepackage[a4paper,margin=1.04in]{geometry} \usepackage[T1]{fontenc} \usepackage{lmodern,amsmath,amssymb,amsthm,mathtools,mathrsfs,microtype} \usepackage{enumitem,booktabs,array} \usepackage[hidelinks]{hyperref} \hypersetup{pdftitle={Factorial Moments in Two Variables: Exact Leaf Transport and Trace Orthogonality},pdfauthor={Christopher D. Long}} \setlength{\emergencystretch}{2em} \numberwithin{equation}{section} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition}\newtheorem{definition}[theorem]{Definition} \theoremstyle{remark}\newtheorem{remark}[theorem]{Remark} \newcommand{\C}{\mathbb C} \newcommand{\R}{\mathbb R} \newcommand{\Q}{\mathbb Q} \newcommand{\Qb}{\overline{\mathbb Q}} \newcommand{\Lfac}{\mathcal L} \newcommand{\CC}{\mathcal C} \newcommand{\HH}{\widehat G} \newcommand{\dd}{\,d} \newcommand{\Crit}{\operatorname{CritVal}} \newcommand{\Span}{\operatorname{span}} \newcommand{\Res}{\operatorname{Res}} \newcommand{\ord}{\operatorname{ord}} \newcommand{\UC}{U_d^{\C}} \newcommand{\GL}{\operatorname{GL}} \newcommand{\one}{\mathbf 1} \title{Factorial Moments in Two Variables:\\ Exact Leaf Transport and Trace Orthogonality} \author{Christopher D. Long\\\small\texttt{galizur@gmail.com}} \date{21 November 2026} \begin{document} \maketitle \begin{abstract} Let $\Lfac(x^ay^b)=a!b!$. We prove factorial-moment nonvanishing for nonzero polynomials in two variables whose highest homogeneous part is not a scalar multiple of $(x+y)^D$. The proof uses an exact radial Mittag--Leffler transform, polynomial monodromy, and Lindemann--Weierstrass. The transform produces vanishing leaf germs containing a pole density and a Volterra correction. The corrections continue with their full integration histories retained and have no negative Puiseux coefficients at a nonzero terminal critical value. The resulting weighted coefficient annihilates the top rational monodromy component, while the unweighted inverse-derivative vector belongs to that component by a zero-trace identity on every proper composition block. Conjugate pairing excludes all nonzero critical values, and endpoint leaf germs dispose of the remaining pure powers. Combining the nonflat theorem with the explicitly stated single-integral nonvanishing theorem of EIT~I gives the two-variable Factorial Conjecture. As a formal consequence, the zero-integral polynomials for the product-exponential measure $e^{-x-y}\dd x\dd y$ form a Mathieu--Zhao subspace; more strongly, a polynomial whose factorial moments vanish for all sufficiently large powers is zero. Thus the two-dimensional product-exponential case of van den Essen's Integral Conjecture follows. The nonflat argument is independent of EIT~I. \end{abstract} \section{The statement and the inputs}\label{sec:introduction} Define the complex-linear functional $\Lfac:\C[x,y]\to\C$ by $\Lfac(x^ay^b)=a!b!$ for $a,b\ge0$. The Factorial Conjecture was introduced by van den Essen, Wright and Zhao~\cite{FC}. Liu and Sun proved the homogeneous two-variable case~\cite{LiuSun}. The present argument treats arbitrary lower homogeneous terms. For $0\ne f\in\C[x,y]$ of total degree $D\ge1$, write \begin{equation}\label{eq:radial-polynomial} P(t,r)=f(rt,r(1-t))=\sum_{j=0}^{D}A_j(t)r^{D-j}, \qquad \deg A_j\le D-j,\qquad A=A_0. \end{equation} Restriction of degree-$D$ homogeneous forms to $x+y=1$ is injective: off $x+y=0$, homogeneity recovers the form from this restriction. Thus $A(t)=f_D(t,1-t)$ is nonzero. Its degree $d$ can be smaller than $D$. We call the cases $d>0$ and $d=0$ \emph{nonflat} and \emph{flat}, respectively. In particular, \[ A\text{ is constant}\quad\Longleftrightarrow\quad f_D(x,y)=c(x+y)^D\quad\text{for some }c\ne0. \] \begin{theorem}[Nonflat nonvanishing]\label{thm:nonflat} If $0\ne f\in\C[x,y]$ has nonconstant $A$, then $\Lfac(f^n)\ne0$ for some $n\ge1$, and in fact for infinitely many $n$. \end{theorem} \begin{theorem}[EIT~I: the external nonvanishing input]\label{thm:eit} Let $q\in\Qb[t]$ be nonconstant, let $a,b\in\Qb$ be distinct, and let $\xi\in\Qb^\times$. Then \[ \int_a^b e^{\xi q(t)}\dd t\ne0. \] \end{theorem} This is the nonvanishing part of the single-integral corollary of EIT~I~\cite[Corollary~1.4, source label \texttt{cor:single}]{EIT}; that result also asserts transcendence. The integral can be taken along any finite path, since its integrand is entire. No extension to arbitrary amplitudes is being used. \begin{corollary}[FC(2), using Theorem~\ref{thm:eit}]\label{cor:fc2} For every $f\in\C[x,y]$, \[ \bigl(\Lfac(f^n)=0\text{ for every }n\ge1\bigr)\quad\Longrightarrow\quad f=0. \] Equivalently, every nonzero $f$ has infinitely many nonzero factorial moments. \end{corollary} The following consequence belongs to the original Image/Mathieu framework in which the Factorial Conjecture was introduced. Recall that a linear subspace $M$ of a commutative algebra $R$ is a Mathieu subspace (now often called a Mathieu--Zhao subspace) if, whenever $f^m\in M$ for every $m\ge1$, one has $gf^m\in M$ for all sufficiently large $m$ for every $g\in R$. Van den Essen's Integral Conjecture predicts this property for the kernel of integration against a positive measure with finite polynomial moments~\cite{AmazingImage}. Thus the implication below is formal rather than a new connection; the substantive new input is Corollary~\ref{cor:fc2}. \begin{corollary}[Two-dimensional product-exponential Integral Conjecture] \label{cor:integral-mathieu} Let \[ B=(0,\infty)^2,\qquad \dd\sigma=e^{-x-y}\dd x\dd y, \] and let \[ M=\left\{h\in\C[x,y]: \int_B h\,\dd\sigma=0\right\}. \] Then $M=\ker\Lfac$ is a Mathieu--Zhao subspace of $\C[x,y]$. In fact, \[ \left\{f\in\C[x,y]: \Lfac(f^m)=0\text{ for all sufficiently large }m\right\}=\{0\}. \] Consequently, the Integral Conjecture holds for the product-exponential measure $e^{-x-y}\dd x\dd y$ in dimension two. \end{corollary} \begin{proof} For monomials, \[ \int_B x^a y^b e^{-x-y}\dd x\dd y=a!b!, \] so $M=\ker\Lfac$. Suppose first that $\Lfac(f^m)=0$ for all $m\ge N$. Then for every $k\ge1$, \[ \Lfac\bigl((f^N)^k\bigr)=\Lfac(f^{Nk})=0. \] Corollary~\ref{cor:fc2} applied to $f^N$ gives $f^N=0$, hence $f=0$. This proves the displayed eventual-power assertion. In particular, if $f^m\in M$ for every $m\ge1$, then $f=0$, and therefore $gf^m=0\in M$ for every $g\in\C[x,y]$ and every $m\ge1$. Thus $M$ is a Mathieu--Zhao subspace. Since $B$ is open, $\sigma$ is positive, and all of its polynomial moments are finite, this is precisely the stated instance of the Integral Conjecture. \end{proof} \paragraph{Direct dependencies and scope.} The nonflat proof uses Lindemann--Weierstrass and the top-component irreducibility theorem for rational permutation modules of groups containing a full cycle~\cite[Theorem~3.1 and Section~4.1]{PM}. These are published inputs. We state precisely the part of that representation theorem needed below, and prove the endpoint averaging, planar separation, and resulting path-module containment in our notation. We do not reprove the underlying Schur-ring classification. For the unrestricted flat conclusion, the additional external input is Theorem~\ref{thm:eit}. Its source is a separate paper published on Hexagon; its proof is not reproduced here. The $E$-function lifting and Fourier--Laplace theory used there are transitive dependencies of that input. Apart from standard complex analysis, algebraic inverse branches, and the Nullstellensatz and L\"uroth theorems, the remaining main-text arguments are included. None of the appendices is a logical input to either main conclusion. \paragraph{Manuscript status.} This is an unrefereed research manuscript, not a Lean formalization. The EIT dependence of Corollary~\ref{cor:fc2} is explicit; the nonflat argument is independent of EIT. The exact checks accompanying the source verify particular identities, not the universal theorem. \section{Normalization and the exact kernel}\label{sec:kernel} \begin{lemma}[Algebraic specialization]\label{lem:specialization} A complex counterexample with specified total degree $D$ and specified positive degree $d$ of $A$ would give a counterexample over $\Qb$ with the same degrees. The same assertion holds with the normalized flat top $A=1$. \end{lemma} \begin{proof} Use the finitely many coefficients of a polynomial of degree at most $D$ as variables. All equations $\Lfac(f^n)=0$ have rational coefficients. Impose the linear conditions specifying $\deg A=d$, and adjoin the inverse of its degree-$d$ coefficient. In the flat case impose $A=1$, after scaling the proposed counterexample by its nonzero constant leading value. Evaluation at the assumed complex point shows that the ideal over $\Qb$ is proper. Noetherianity and the Nullstellensatz give an algebraic point annihilating it. Every moment polynomial belongs to the ideal, so this point satisfies the entire family of moment equations. The inverse equation ensures that the degree-$d$ coefficient of $A$ is nonzero; hence $\deg A=d$ and $f_D\ne0$, preserving total degree $D$. In the flat case $A=1$ likewise ensures $f_D\ne0$. Neither the monodromy group nor the decomposition lattice has to survive specialization, since the contradiction is uniform in both. \end{proof} Set \begin{align} M_n&=\frac{\Lfac(f^n)}{\Gamma(Dn+2)}, & \HH(w)&=\frac1w\sum_{n\ge0}M_nw^{-n},\label{eq:normalized}\\ E_{D,2}(z)&=\sum_{n\ge0}\frac{z^n}{\Gamma(Dn+2)}, & \CC(w,t)&=\frac1w\int_0^\infty e^{-r}r E_{D,2}\!\left(\frac{P(t,r)}w\right)\dd r.\label{eq:kernel} \end{align} Termwise integration of a polynomial gives \[ \Lfac(f^n)=\int_0^\infty\!\int_0^\infty e^{-x-y}f(x,y)^n\dd x\dd y =\int_0^1\!\int_0^\infty e^{-r}rP(t,r)^n\dd r\dd t. \] The last equality uses the Jacobian $r$ of $x=rt$, $y=r(1-t)$. For sufficiently large $|w|$ the resulting generating integral is normally convergent, and \begin{equation}\label{eq:full-transform} \HH(w)=\int_0^1\CC(w,t)\dd t. \end{equation} Indeed, for $x=|z|^{1/D}$ the defining series gives the global bound \begin{equation}\label{eq:ML-growth} |E_{D,2}(z)|\le\sum_{n\ge0}\frac{x^{Dn}}{(Dn+1)!} \le\sum_{m\ge0}\frac{x^m}{m!}=e^x. \end{equation} Since $|P(t,r)|^{1/D}\le C(1+r)$ on a compact $t$-set, this estimate also justifies summation under the integral for sufficiently large $|w|$. All positive moments vanish exactly when $\HH(w)=1/w$ near infinity. Put $V=\{t:A(t)\ne0\}$. We need a coefficient on a finite cover, not a single-valued logarithmic coefficient on the entire base. \begin{lemma}[Exact local kernel and a continued coefficient]\label{lem:kernel} Near $w=A(t)\ne0$, a continued germ of the kernel has the form \begin{equation}\label{eq:local-kernel} \CC(w,t)=\frac{E(t)}{w-A(t)}+ B_0(w,t)\log(w-A(t))+H(w,t),\qquad E(t)=\exp\!\left(\frac{A_1(t)}{DA(t)}\right), \end{equation} where $B_0,H$ are locally jointly holomorphic. The residue is unchanged by continuation. For fixed $t\in V$, the only possible finite exceptional values for continuation in $w$ are $0,A(t)$, locally uniformly in $t$. The distinguished diagonal germ of $B_0$ extends to a holomorphic function $\mathscr B$ on \begin{equation}\label{eq:cover} \mathscr X=\{(w,t,\rho):w\ne0,\ t\in V,\ \rho^D=A(t)/w\}. \end{equation} It is given by \begin{equation}\label{eq:global-B} \mathscr B(w,t,\rho)=\frac{\rho^{D-1}}{DA(t)} \sum_{m\ge1}\frac{(-1)^mF_m(t)}{(m-1)!}(1-\rho)^{m-1}, \end{equation} where $F_m$ are defined in \eqref{eq:Laurent}. The series is entire in $1-\rho$, locally uniformly on $V$. The initial diagonal germ is the restriction with $\rho$ near $1$. \end{lemma} \begin{proof} Let $\omega=e^{2\pi i/D}$. Expansion of the exponentials gives the exact root filter \begin{equation}\label{eq:filter} E_{D,2}(X^D)=\frac1{DX}\sum_{k=0}^{D-1}\omega^{-k}e^{\omega^kX} -C_{D,1}(X),\quad C_{D,1}=\begin{cases}X^{-1},&D=1,\\0,&D\ge2.\end{cases} \end{equation} Fix a compact set $K\subset V$. Split the radial integral at a sufficiently large $R$, chosen uniformly for $t\in K$, and apply the filter only to the exterior part. Leave the compact part in its entire Mittag--Leffler form. Initially impose $|w|>|A(t)|$, so $|\rho|=|A(t)/w|^{1/D}<1$ and all $D$ exponential summands converge. Each summand is then continued separately from this common domain. On $V$, the exterior root with leading term $r$ and its inverse are \[ \Phi_t(r)=(P(t,r)/A(t))^{1/D}=r+p_1(t)+O(r^{-1}),\quad \chi(t,Y)=Y-p_1(t)+O(Y^{-1}),\quad p_1=\frac{A_1}{DA}. \] The analytic inverse theorem at infinity gives convergent Laurent series on common exterior disks for $t\in K$; one first works on a neighborhood of $K$ and then uses compactness. The chosen leading terms specify the roots and inverses uniquely, so their coefficients are single-valued and holomorphic on $V$. Define \begin{equation}\label{eq:Laurent} F(t,Y)=e^{Y-\chi(t,Y)}\frac{\chi(t,Y)\chi_Y(t,Y)}Y =\sum_{m\ge0}F_m(t)Y^{-m}. \end{equation} Here $F_0=E$. By a uniform Cauchy estimate at infinity, there are $C_K,R_0>0$ such that $|F_m(t)|\le C_KR_0^m$ for every $t\in K$ and $m\ge0$. Enlarge $R$ so that the image paths lie in this common exterior domain. With $\rho=(A(t)/w)^{1/D}$ near $1$, the summand singular at this diagonal is \begin{equation}\label{eq:singular-summand} \frac{\rho^{D-1}}{DA(t)} \int_{\Phi_t([R,\infty))}e^{-(1-\rho)Y}F(t,Y)\dd Y. \end{equation} In $\Re(1-\rho)>0$ the exterior image ray can be moved to a common positive ray. The far connector tends to zero exponentially, and the finite connector is entire in $1-\rho$, locally uniformly in $t$. Choose a starting point $R_1>R_0$ on that ray and put \[ J_m(\eta)=\int_{R_1}^\infty e^{-\eta Y}Y^{-m}\dd Y. \] Direct integration and integration by parts give \begin{align*} J_0&=e^{-\eta R_1}/\eta,\\ J_1&=-\log\eta-\gamma-\log R_1- \sum_{j\ge1}\frac{(-\eta R_1)^j}{j\,j!},\\ J_{m+1}&=\frac{e^{-\eta R_1}R_1^{-m}-\eta J_m}{m}\qquad(m\ge1). \end{align*} The logarithmic coefficient of $J_m$ is $(-1)^m\eta^{m-1}/(m-1)!$. For $m\ge1$ put \[ H_m(\eta)=J_m(\eta) -\frac{(-1)^m\eta^{m-1}}{(m-1)!}\log\eta. \] The displayed formulas show that $H_m$ is entire and obeys the same recurrence as $J_m$. On $|\eta|\le L$, set $B_m=R_1^{m-1}\sup|H_m|$. Then \[ B_{m+1}\le\frac{e^{LR_1}}m+\frac{LR_1}m B_m. \] For $m\ge2LR_1$ the last coefficient is at most $1/2$; enlarging a constant to cover the finitely many earlier terms proves $\sup_{|\eta|\le L}|H_m|\le C_{L,R_1}R_1^{1-m}$. Together with $R_1>R_0$ and the bound on $F_m$, this proves normal convergence of the entire remainders. The logarithmic coefficients converge normally by their factorial denominators. Thus the tail is \[ \frac{E(t)}\eta+ \left(\sum_{m\ge1}\frac{(-1)^mF_m(t)}{(m-1)!}\eta^{m-1}\right)\log\eta +\text{a function holomorphic near }\eta=0. \] This supplies exact continuation on the universal cover of $\eta\ne0$. Since \[ 1-\rho=\frac{w-A(t)}{DA(t)}+O((w-A(t))^2), \] \eqref{eq:singular-summand} has residue $E(t)$ and logarithmic coefficient \eqref{eq:global-B}. Changing $\log(1-\rho)$ to $\log(w-A(t))$ changes only a holomorphic remainder. The entireness used below also has a contour formulation. For any $R_2>R_0$, orient $|Y|=R_2$ counterclockwise. Termwise integration gives \[ \Lambda(t,\eta):=\sum_{m\ge1}\frac{(-1)^mF_m(t)}{(m-1)!}\eta^{m-1} =-\frac1{2\pi i}\int_{|Y|=R_2}e^{-\eta Y}F(t,Y)\dd Y. \] Thus $|\Lambda(t,\eta)|\le C_{K,R_2}e^{R_2|\eta|}$, uniformly on $K$, and $\mathscr B=\rho^{D-1}\Lambda(t,1-\rho)/(DA(t))$ is holomorphic on the stated cover. This contour formula is for $\Lambda$; the prefactor must be retained when recovering $\mathscr B$. The other root summands have parameters $1-\omega^k\rho$; their zeros all project to $w=A(t)$. At $\rho=1$ they are locally holomorphic for $k\ne0$. The compact part and the $D=1$ correction are holomorphic at every nonzero diagonal. This proves the claimed exceptional set and uniformity. A change of logarithm branch adds a function holomorphic on the cover, and cannot alter the pole residue. \end{proof} \begin{remark}[Canonical local coefficient versus continued germ]\label{rem:B-sheets} The local logarithmic coefficient of the complete kernel is canonical at its nonzero diagonal. The continuation of its distinguished germ, however, is $\mathscr B(w,t,\rho)$ on $\mathscr X$ and need not descend to the base $(w,t)$-space. If another root summand becomes singular, the identity $\omega^{-k}\rho^{D-1}=(\omega^k\rho)^{D-1}$ gives the same canonical local coefficient after reselecting that summand. It does not identify different continued root sheets. Every integral below retains its root parameter. Appendix~\ref{app:examples} gives an exact sign-changing example. \end{remark} \begin{lemma}[Normality near zeros of the top]\label{lem:normality} Near a zero of $A$, the original kernel integral is normally convergent for $w$ on any fixed compact set away from zero. On every bounded $t$-disk it is holomorphic in $t$ for all sufficiently large $|w|$. \end{lemma} \begin{proof} For $D>1$, locally uniformly, \[ |P(t,r)/w|^{1/D}\le |A(t)/w|^{1/D}r+C(1+r)^{1-1/D}. \] For $D=1$ the second term is bounded. Near a zero of $A$, or for $|w|$ sufficiently large on a bounded $t$-disk, make the first coefficient less than $1/2$. By \eqref{eq:ML-growth}, on compact $w$-sets away from zero the kernel integrand is bounded by \[ C_1r\exp\bigl(-r/2+C_2(1+r)^{1-1/D}\bigr), \] with the last contribution bounded when $D=1$. This is integrable and gives joint holomorphy locally in $(w,t)$, including at zeros of $A$. To compare with a continued germ, first fix the entire finite continuation path in $w$ away from zero. Its modulus has a positive minimum, so near a zero of $A$ one can shrink the $t$-neighborhood until $|A(t)/w|^{1/D}<1/2$ along the whole path. The single-valued integral agrees with the germ near infinity, hence with its continuation along this path by uniqueness. No uniform claim over an unbounded $t$-set is needed. \end{proof} \section{Leaf identities and continuation with retained paths}\label{sec:transport} Choose a regular value $b$ different from $0,A(0),A(1)$. Let $T$ be an embedded finite star centered at $b$, with one arm to each distinct member of \[ \{0,A(0),A(1)\}\cup\Crit(A). \] The arms have disjoint interiors and are straight near their leaves. The inverse graph $\Gamma=A^{-1}(T)$ is a tree. Here is a useful direct justification. If $T$ has $s$ arms, then $\Gamma$ has $ds$ edges, $d$ center vertices, and $ds-(d-1)$ leaf vertices: the total finite ramification of a degree-$d$ polynomial is $d-1$. It is connected because the local monodromies about the leaf values generate a transitive covering group; their cycles join precisely the incident lifted stars. Thus its Euler characteristic is $1$, and it is a tree. This is the usual polynomial inverse-tree construction~\cite[Section~2.1]{PM}. Let $\mu$ be its unique simple path from the actual point $0$ to the actual point $1$. Both $\mu$ and $[0,1]$ lie in a bounded disk. Lemma~\ref{lem:normality} therefore permits replacing the interval by $\mu$ in \eqref{eq:full-transform} for large $w$. Append a simple arc from $b$ to infinity, disjoint from $T$ except at $b$ and straight sufficiently near infinity. Label the inverse branches over $b$ so that the geometric monodromy obtained by following this arc, making one counterclockwise turn on a large circle, and returning is $g_\infty=(1\ 2\ \cdots\ d)$. We retain this geometric labeling throughout; an arbitrary generator of the same cyclic subgroup need not preserve the circular order used below. Orient each target arm from its leaf $v$ toward $b$. The path gives a vector $n_v\in\{0,1,-1\}^d$, recording the signed use of the lift in each sheet. With these orientations, \begin{equation}\label{eq:tree} \HH(w)=\sum_v\int_v^b\sum_{i=1}^d n_{v,i} \CC(w,t_i(u))t_i'(u)\dd u. \end{equation} The complement $\widehat{\C}\setminus T$ in the Riemann sphere is simply connected. For $t\in\mu$, the possible kernel singularities $0,A(t)$ lie in $T$, and the kernel is holomorphic at infinity. Lemmas~\ref{lem:kernel} and~\ref{lem:normality} therefore determine its single-valued continuation off $T$, locally uniformly along the compact path $\mu$. Consequently \eqref{eq:tree} holds there by continuation from infinity. It retains every lift over each leaf. Fix $v\ne0$ and write $c_i=\lim_{u\to v}t_i(u)$ on its arm. For $w$ near $v$, define the germ \begin{equation}\label{eq:leaf-germ} \mathscr L_v(w)=\sum_i n_{v,i} \left\{E(t_i(w))t_i'(w)+ \int_{c_i}^{t_i(w)}\mathscr B(w,t,\rho)\dd t\right\}. \end{equation} Initially the integrals are local near $c_i$. Their lifts to $\mathscr X_w$ are selected by $\rho=1$ at the upper endpoint. Since $A(c_i)=v\ne0$, regular and ramified endpoints are both allowed. \begin{lemma}[Complete local singular extraction]\label{lem:local-extraction} If $\HH$ is rational, then $\mathscr L_v\equiv0$ for every nonzero leaf $v$. \end{lemma} \begin{proof} Only meromorphy of the chosen continued germ of $\HH$ at $v$ is used in this local argument. Put $z=w-v$ and $x=u-v$. The germ arriving from the complement of $T$ uses the diagonal root $\rho\to1$; its local logarithm may differ from another choice only by a holomorphic term. Near the leaf, set \begin{align*} a(x)&=\sum_i n_{v,i}E(t_i(v+x))t_i'(v+x),\\ b_0(z,x)&=\sum_i n_{v,i} \mathscr B(v+z,t_i(v+x),\rho)t_i'(v+x), \qquad \rho^D=(v+x)/(v+z),\quad\rho\sim1. \end{align*} These have convergent Puiseux expansions in $x$ with a common finite denominator; $b_0$ is also holomorphic in $z$. Every $x$-exponent is greater than $-1$, since a local inverse of degree $k$ has derivative $O(x^{1/k-1})$. All other edges in \eqref{eq:tree} are holomorphic in $w$ near $v$ on the chosen continuation. This follows from Lemma~\ref{lem:kernel} where $A(t)\ne0$, and from Lemma~\ref{lem:normality} near zero top values. The integrated local holomorphic remainder is holomorphic as well. After straightening the short arm, its relevant part is \[ \int_0^\epsilon \left(\frac{a(x)}{z-x}+b_0(z,x)\log(z-x)\right)\dd x. \] A common nonzero linear rescaling of $x,z$ does not affect the following vanishing argument. Take $z$ in a disk of radius less than $\epsilon$, slit along its positive radius. Then $z-x\ne0$ for $0\le x\le\epsilon$, and $\log(z-x)$ has a continuous choice on this product, holomorphic in $z$; at $x=\epsilon$ it is holomorphic on the full smaller disk. Put $q(z,x)=\int_0^x b_0(z,s)\dd s$. This tends to zero at the lower endpoint. Integration by parts gives a holomorphic boundary term plus \begin{equation}\label{eq:corrected-cauchy} \int_0^\epsilon\frac{a(x)+q(z,x)}{z-x}\dd x. \end{equation} In particular the sign of the correction is positive. There are finitely many fractional classes. Choose their representatives $-1<\delta\le0$ and write \[ a(x)+q(z,x)=\sum_\delta x^\delta h_\delta(z,x), \] where each $h_\delta$ is holomorphic in the ordinary variables $z,x$. Divide $h_\delta(z,x)=h_\delta(z,z)+(x-z)k_\delta(z,x)$. The $k_\delta$-term in \eqref{eq:corrected-cauchy} is holomorphic in $z$, by integrability of $x^\delta$. The only nonholomorphic term from this class is therefore \[ h_\delta(z,z)I_\delta(z),\qquad I_\delta(z)=\int_0^\epsilon\frac{x^\delta}{z-x}\dd x. \] For $-1<\delta<0$, the singular part of $I_\delta$ is a nonzero multiple of $z^\delta$ on a fixed slit; scaling and the beta integral give a factor proportional to $\pi/\sin(\pi\delta)$. For $\delta=0$ it is $\log z$, up to a holomorphic function. Let $T_0$ denote continuation once around $z=0$, and choose $N$ divisible by all Puiseux denominators. Applying $T_0^N-1$ to a meromorphic germ kills the fractional terms and isolates $2\pi iN h_0(z,z)$, so $h_0(z,z)=0$. The remaining terms $z^\delta h_\delta(z,z)$ have distinct monodromy eigenvalues $e^{2\pi i\delta}\ne1$. The corresponding polynomial projectors in $T_0$ isolate each term from the others and from the meromorphic part. Since the constants multiplying $z^\delta$ are nonzero, every $h_\delta(z,z)$ vanishes. Hence \[ a(z)+\int_0^z b_0(z,x)\dd x=0. \] The change of variables $t=t_i(v+x)$ identifies this with \eqref{eq:leaf-germ}. This proves vanishing of the whole corrected germ, not only of its leading coefficient. \end{proof} \begin{remark}[The triangular coefficient equations]\label{rem:triangular} If $a(x)=\sum a_\alpha x^\alpha$ and $b_0(z,x)=\sum b_{p,\beta}z^p x^\beta$, the preceding identity is equivalent to \begin{equation}\label{eq:triangular} a_\alpha+\sum_{p+\beta+1=\alpha}\frac{b_{p,\beta}}{\beta+1}=0 \qquad(\alpha>-1). \end{equation} It is not an identity for $a$ alone. For example, $-\sqrt{x}/(z-x)+(2\sqrt{x})^{-1}\log(z-x)$ is $\partial_x(\sqrt{x}\log(z-x))$, and its integral is holomorphic near $z=0$, although both densities are nonzero. \end{remark} Write $G_A$ for the inverse-branch monodromy group. Fix a terminal path from $b$ toward a nonzero critical value $\lambda$, with a final straight segment, avoiding zero and all critical values before its endpoint. For any negative Puiseux exponent $\alpha$, define the linear functional \begin{equation}\label{eq:coefficient-functional} \Phi_{\lambda,\alpha}(e_i)= [(w-\lambda)^\alpha] \frac{\exp(A_1(t_i(w))/(Dw))}{A'(t_i(w))} \end{equation} using continuation along that path and a common Puiseux parameter. \begin{theorem}[Transport of the full leaf identity]\label{thm:transport} If $\HH$ is rational, $\Phi_{\lambda,\alpha}$ annihilates $\Q[G_A]n_v$ for every nonzero leaf $v$ and every $\alpha<0$. \end{theorem} \begin{proof} Let $U=\C\setminus(\{0\}\cup\Crit(A))$. Continue the zero germ $\mathscr L_v$ along its arm to $b$, and then along any path in $U$. The initial local integration paths lie in $V$, and every moving inverse branch stays there because $A(t_i(w))=w\ne0$. For an individual integral in \eqref{eq:leaf-germ}, retain its path from the fixed $c_i$ and append the path of its moving upper endpoint. Lift the accumulated path to $\mathscr X_w$ with $\rho=1$ at that upper endpoint. The fixed lower root is allowed to change after a loop; it remains part of the data. Here is a precise local description. On a small $w$-disk $W$ away from zero choose a holomorphic root $s(w)^D=w$, and put \[ \Sigma=\{(t,\sigma):t\in V,\ \sigma^D=A(t)\}, \qquad \rho=\sigma/s(w). \] Then $\mathscr X|_W$ is identified with $W\times\Sigma$. The curve $\Sigma$ is locally parametrized by $t$, since $\sigma\ne0$, even at a critical point of $A$. The differential $\mathscr B(w,t,\sigma/s(w))\dd t$ is holomorphic on this product. A fixed compact lifted path integrates to a holomorphic function of $w$; a short appended piece has a jointly holomorphic local primitive. This constructs the continuation along each compact portion of the chosen path. At the moving endpoint $\sigma=s(w)$, so $\rho=1$ is preserved. Different roots at a fixed lower endpoint are never reset. Now approach $\lambda\ne0$, and suppose an endpoint tends to $c$ with $A(c)=\lambda$. Split the accumulated history into a fixed compact lifted path ending near $c$, and a short terminal piece. The first part is holomorphic on a disk about $\lambda$ by the product description. A point with $A(t)=w$ on that history causes no singularity of $\mathscr B$, because its coefficient series is entire in $1-\rho$. No extension across $A(t)=0$ is asserted or used. Near $(\lambda,c)$ the root with $\rho\sim1$ is holomorphic. A primitive $J(w,t)$ of the corresponding differential is jointly holomorphic, and the terminal piece is \[ J(w,t_i(w))-J(w,t_*). \] If $A$ has local degree $k$ at $c$, then $t_i(w)=c+O((w-\lambda)^{1/k})$. Thus this primitive has only nonnegative Puiseux exponents. The same holds for every retained history. Every negative coefficient of the continued identity comes solely from $E(t_i(w))t_i'(w)$. For each $g\in G_A$, choose a loop at $b$ realizing $g$, and then follow the fixed terminal path. The continued $\mathscr L_v$ is still zero, so its negative coefficient is $\Phi_{\lambda,\alpha}(g n_v)=0$. Linear extension proves the theorem. Removing the extra regular point zero from the base, when zero is not critical, does not remove any permutation from its monodromy image. The complete histories can depend on more than $g$; no action of $G_A$ on those histories is asserted. Only the negative Puiseux coefficients factor through the permutation action: the history terms contribute none, while the pole contribution depends only on $g$. \end{proof} \section{The path module and the simultaneous trace identity}\label{sec:module} We use the bilinear pairing $\langle x,y\rangle=\sum_i x_i y_i$. Vector spaces are over $\Q$ until explicitly complexified. For the actual signed vectors in \eqref{eq:tree}, put \[ M=\sum_v\Q[G_A]n_v,\qquad M_{\ne0}=\sum_{v\ne0}\Q[G_A]n_v. \] \begin{lemma}[The zero leaf is algebraically redundant]\label{lem:zero-leaf} One has $\sum_v n_v=0$ coordinatewise, and $M_{\ne0}=M$. \end{lemma} \begin{proof} Each visited center vertex of $\mu$ has one entering and one leaving edge, contributing opposite signs to its sheet coordinate. Neither actual endpoint lies over $b$. Thus $n_0=-\sum_{v\ne0}n_v$; apply all monodromy elements. This makes no analytic assertion at $w=0$. \end{proof} Use the geometric labeling of Section~\ref{sec:transport}, with $g_\infty=(1\ 2\ \cdots\ d)$, and identify sheet $i$ with $i-1$ in $\mathbb Z/d\mathbb Z$. For $k\mid d$, let $V_k\subset\Q^d$ consist of $k$-periodic vectors, and set \begin{equation}\label{eq:top-component} D(G_A)=\{k\mid d:V_k\text{ is }G_A\text{-invariant}\},\qquad U_d=\bigcap_{\substack{k\in D(G_A)\\kn$. The inequalities $K\le J\le DK$ show $0\le C_{n,\mathbf k}\le1$. Indeed, the smallest $K$ factors in the denominator product dominate the numerator factors: \[ Dn-J+2+j\ \ge\ D(n-K)+2+j\ \ge\ n-K+1+j \qquad(0\le j0. \] This is summable over $\mathbb N^D$ and proves uniform dominated convergence, including $D=1$. \end{proof} \begin{proof}[Proof of Corollary~\ref{cor:fc2}] The nonflat case is Theorem~\ref{thm:nonflat}, and a nonzero constant polynomial is immediate. Suppose $D\ge1$ and $A$ is constant. It is nonzero, so scaling $f$ permits $A=1$. If all positive moments vanish, specialize algebraically with this normalization retained. Equation~\eqref{eq:radial-limit} gives \[ \lim_{n\to\infty}M_n=\int_0^1\exp\!\left(\frac{A_1(t)}D\right)\dd t. \] When $A_1$ is constant this limit is a nonzero exponential. Otherwise Theorem~\ref{thm:eit} applies with $q=A_1$, $a=0$, $b=1$ and $\xi=1/D$. Its hypotheses hold over $\Qb$, and the amplitude is exactly $\dd t$. The limit is again nonzero, contradicting $M_n=0$ for every $n\ge1$. The infinitely-many assertion follows by applying the result to $f^N$ if a tail of its factorial moments were zero. \end{proof} \paragraph{An elementary flat subcase.} The EIT input is unnecessary when $A$ is constant and $\deg A_1\le1$. After normalizing $A=1$, retain the condition $\deg A_1\le1$ as rational linear equations in the specialization argument. Write $A_1(t)/D=\alpha t+\beta$ over $\Qb$. Then \[ \int_0^1e^{\alpha t+\beta}\dd t =\begin{cases} e^\beta,&\alpha=0,\\[2pt] e^\beta(e^\alpha-1)/\alpha,&\alpha\ne0. \end{cases} \] This is nonzero by Hermite--Lindemann~\cite{Baker}. Thus the same radial limit proves nonvanishing in this subcase using published inputs only; in particular it handles the flat case in total degree $D\le2$. No rate of convergence beyond the uniform $o(1)$ in Lemma~\ref{lem:radial-limit} is needed. \paragraph{Research provenance and assistance.} During the development of this argument, the \texttt{openai/math} release suggested examining local jet separation, differential-system multiplicity, and continuation of complete branch data \cite{OAIJets,OAIMultiplicity,OAIContinuation}. These were prompts for proof exploration, not theorems transferred into the FC(2) argument. In particular, local \'etale rigidity does not imply injectivity of a sum over sheets, and the cited billiard continuation results are not applied here. The main proof uses none of the release's results. Appendix~\ref{app:certificates} records the separate finite-jet application of Moura's classical estimate, together with its formal proof in the release. ChatGPT assisted with exploration, drafting, symbolic checking and review; the author is responsible for the mathematical claims and their attribution. \appendix \section{Exact continuation examples}\label{app:examples} The examples in this appendix illustrate the continuation conventions. None is a logical input to the main proof in Sections~\ref{sec:kernel}--\ref{sec:flat}. \subsection{The pole alone has the wrong jump} For $f=3x+2y+1$, one has $D=1$ and $A(t)=2+t$. Its exponential generating series is \[ \sum_{n\ge0}\frac{\Lfac(f^n)}{n!}z^n =\frac{e^z}{(1-3z)(1-2z)}. \] Integrating the series and substituting $z=1/w$ gives the exact identities \[ \HH(w)=\int_0^{1/w}\frac{e^z}{(1-3z)(1-2z)}\dd z, \qquad \HH'(w)=-\frac{e^{1/w}}{(w-3)(w-2)}. \] Counterclockwise continuation around $3$ therefore gives $\Delta_3\HH=-2\pi i e^{1/3}$. In contrast, the pole integral alone is \[ C_0(w)=\int_2^3\frac{e^{1/u}}{w-u}\dd u, \qquad \Delta_3 C_0=-2\pi i e^{1/w}. \] The difference is a nonconstant analytic germ. For this family, $\mathscr B(w,t)=e^{1/A(t)}/A(t)^2$, and the corrected branch germ at a lower endpoint with value $v$ is \[ e^{1/w}+\int_v^w\frac{e^{1/u}}{u^2}\dd u=e^{1/v}. \] Thus the correction, while one local singular order softer, changes the entire jump. A leading-pole identity cannot be substituted for \eqref{eq:leaf-germ}. \subsection{A coefficient that changes root sheets} Take $D=2$ and $P=Ar^2+c$, where $c\ne0$ and $A=A(t)\ne0$. The coefficient formula can be computed independently of radial reversion. On writing $n=m+k$, the duplication formula gives \[ \binom{m+k}{k}\frac{\Gamma(2m+2)}{\Gamma(2m+2k+2)} =\frac{4^{-k}\Gamma(m+3/2)}{k!\,\Gamma(m+k+3/2)}. \] For $k\ge1$, use the beta integral for the ratio and sum the geometric series over $m$. For large $|w|$ this gives \begin{equation}\label{eq:quadratic-beta} \CC(w,t)=\frac1{w-A} +\int_0^1\frac{u^{1/2}}{w-Au} \sum_{k\ge1}\frac{(c/(4w))^k(1-u)^{k-1}}{k!(k-1)!}\dd u. \end{equation} All exchanges converge normally there. At the endpoint $u=1$, the logarithmic coefficient is minus $1/A$ times the analytic numerator evaluated at $u=w/A$. Reindexing $k=j+1$ yields \begin{equation}\label{eq:quadratic-B} \mathscr B(w,t,\rho) =-\frac{c\rho}{4A^2} \sum_{j\ge0}\frac{[c(\rho^2-1)/(4A)]^j}{j!(j+1)!}, \qquad \rho^2=A/w. \end{equation} This is an all-orders convergent identity. In particular, $\mathscr B(A,t,1)=-c/(4A^2)$, agreeing with \eqref{eq:global-B}. The right side of \eqref{eq:quadratic-B} is odd in $\rho$. A fixed-$t$ loop around zero changes the sign of the continued distinguished coefficient. When the loop returns to a neighborhood of the diagonal, the canonical coefficient of the complete kernel is unchanged because the singular root-filter summand has been reselected. This distinguishes the two objects in Remark~\ref{rem:B-sheets} concretely. \subsection{A moving endpoint with a changed lower root} Let \[ P=(L(t)r+c)^D,\qquad L(t)=at+b,\qquad ac\ne0,\quad D\ge2. \] This is the radial polynomial of $[ax+b(x+y)+c]^D$. Fix a lower endpoint $p$ with $L(p)\ne0$, and write $s=L(t_i(w))$, so $s^D=w$. On the lifted integration path, \[ \Phi=r+c/L,\qquad \chi=Y-c/L,\qquad F=e^{c/L}(1-c/(LY)),\qquad \rho=L/s. \] Formula~\eqref{eq:global-B} now reads \[ \mathscr B\dd t=-\frac{\dd_t(e^{c/L(t)})}{Da\,s^{D-1}}. \] Since $A'(t_i)=Da\,s^{D-1}$, the full corrected expression is \begin{equation}\label{eq:power-history} E(t_i(w))t_i'(w)+\int_p^{t_i(w)}\mathscr B\dd t =\frac{e^{c/L(p)}}{Da\,s^{D-1}}. \end{equation} For a positive loop around zero, $s$ changes to $\omega s$, where $\omega=e^{2\pi i/D}$. The lower root parameter changes to $\omega^{-1}\rho$, and \eqref{eq:power-history} changes by the factor $\omega$. Resetting the lower root, or deleting the integral, would both give the wrong continuation. The general transport proof retains exactly these data. \section{Finite certificates for branch densities}\label{app:certificates} This appendix concerns auxiliary differential systems and is not an input to the main proof. In particular, its branch-density tests do not bound the first nonzero factorial moment. The analytic multiplicity estimate is due to Moura~\cite{Moura,MouraHyper}. We prove the precise polynomial-presentation, formal-power-series version below; the proof is adapted from the version in~\cite{OAIMultiplicity}, which also gives a Lean statement. The formal Wronskian step is explained in Bostan--Dumas~\cite{BostanDumas}. These attributions concern the auxiliary estimate, not a formal verification of FC(2). \subsection{The classical multiplicity estimate} \begin{theorem}[Moura's estimate in a polynomial presentation] \label{thm:multiplicity} Let $F$ have characteristic zero, $m,n\ge1$, $q\in F[z]$ with $q(0)\ne0$ and $\deg q\le n$, and $P\in M_m(F[z])$ with $\deg P_{ij}\le n-1$. Suppose \[ G\in\GL_m(F[[z]]),\qquad qG'=PG. \] For a constant row $u$ and constant column $v$, a nonzero series $y=uGv$ satisfies \begin{equation}\label{eq:moura-bound} \ord_0 y\le(n-1)\binom m2+m-1. \end{equation} \end{theorem} \begin{proof} Put $h=uG$, and let $V$ be the $F$-span of its component series. Its dimension $k$ is between $1$ and $m$. Gaussian elimination on successive coefficients gives a basis $b_1,\ldots,b_k$ of $V$ with strictly increasing orders $a_1<\cdots