\documentclass[11pt,a4paper]{article} \usepackage[utf8]{inputenc} \usepackage[T1]{fontenc} \usepackage{amsmath,amssymb,amsthm} \usepackage[margin=2.6cm]{geometry} \usepackage[hidelinks]{hyperref} \frenchspacing \theoremstyle{plain} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \newtheorem*{thmA}{Theorem A} \newtheorem*{thmC}{Theorem C} \newtheorem*{mainthm}{Main Theorem} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\F}{\mathbb{F}} \newcommand{\rk}{\operatorname{rank}} \newcommand{\cork}{\operatorname{corank}} \newcommand{\Fix}{\operatorname{Fix}} \newcommand{\Img}{\operatorname{im}} \newcommand{\AND}{\mathbin{\&}} \newcommand{\sub}{\subseteq} \newcommand{\GL}{\mathrm{GL}} \newcommand{\old}{\mathrm{old}} \newcommand{\new}{\mathrm{new}} \title{Old and new Drinfeld cusp forms of level $t$ in characteristic $2$:\\ the weights $k=2^m+1$} \author{David Niedbala Giraudin\\[2pt] \small Independent researcher, France\quad ORCID 0009-0009-1526-1178} \date{18 November 2026} \begin{document} \maketitle \begin{abstract} Let $q=2$, $A=\F_2[t]$, and let $U_t$ be the Atkin operator on the space $S_k(\Gamma_0(t))$ of Drinfeld cusp forms of level $t$. For every weight $k=2^m+1$ we prove that $S_k(\Gamma_0(t))$ is the direct sum of its oldforms and its newforms, that $k/2$ is the largest slope of $U_t$, and that no oldform has slope $k/2$. This is Conjecture~1.2(3) of Bandini and Valentino, and their conjectured old/new decomposition, for this infinite family of weights in characteristic $2$, where the known methods do not apply. We also obtain the characteristic polynomial of $U_t$ in closed form in terms of that of the Hecke operator $T_t$ in level one, and the bound: every slope of $T_t$ on $S_k(\GL_2(A))$ is at most $J_m=(2^m-(-1)^m)/3$. The proof rests on a combinatorial theorem: the leading principal minors of the Bandini--Valentino matrix are nonzero exactly at the Jacobsthal numbers; equivalently, the number of perfect matchings of the graph of nonempty subsets of $\{1,\dots,m\}$ joined when disjoint or nested is odd exactly when the number of vertices plus one is a Jacobsthal number. \end{abstract} \paragraph{Status.} Statements labelled \textsc{proved} carry a complete proof in this text. Statements labelled \textsc{verified} have been checked by computer only, over the range indicated; statements labelled \textsc{open} are neither. Every computational assertion is reproduced by the ancillary script \texttt{anc/verif\_all.py} (pure Python with numpy, 52 checks, about one minute). The text has not been read by any person other than the author. Prepared with Claude (Anthropic), which carried out the computations and drafted the proofs under the author's direction. The author is responsible for the content. \section{Introduction} \subsection{Setting} Let $q$ be a power of a prime $p$ and $A=\F_q[t]$. Bandini and Valentino~\cite{BV18,BV19} gave an explicit formula for the Atkin operator $U_t$ on Drinfeld cusp forms of level $\Gamma_0(t)$, and in~\cite{BV22a} they extracted from it a matrix $M$ over $\F_q$ such that, on the block of dimension $n$, \[ U_t=MD,\qquad D=\mathrm{diag}(t^{s_1},\dots,t^{s_n}),\qquad s_i=j+1+(i-1)(q-1),\quad k=2j+2+(n-1)(q-1) \] \cite[(11), (12)]{BV22a}. For $q=2$ one has $j=0$, hence $s_i=i$ and $k=n+1$: the whole space $S_k(\Gamma_0(t))$ has dimension $n=k-1$. Throughout, \[ q=2,\qquad n=2^m,\qquad k=2^m+1\qquad(m\ge1). \] We write $J_r=(2^r-(-1)^r)/3$ for the Jacobsthal numbers $0,1,1,3,5,11,21,43,85,171,\dots$; they satisfy \begin{equation}\label{eq:jac} J_r+J_{r+1}=2^r,\qquad J_{r+1}=J_r+2J_{r-1}. \end{equation} \subsection{Known results} \begin{itemize} \item[(K1)] Every newform of level $t$ has slope $k/2$ \cite[Section~2.6.2]{BV22a}: the Fricke involution gives $U_t^2=t^k$ on the space of newforms. \item[(K2)] Bandini and Valentino conjectured that $S_k(\Gamma_0(t))$ is the direct sum of oldforms and newforms \cite[Conj.~2.6(3)]{BV22a}, and that $k/2$ is the maximal slope, with no oldform of slope $k/2$ \cite[Conj.~1.2(3)]{BV23b}. These are equivalent \cite[Thm.~3.15]{Val22}; and \cite[Thm.~5.1, Rem.~5.3]{BV22a} reduces them to the absence of an eigenvalue $t^{k/2}$ (up to sign) for the Hecke operator $T_t$ in level one. Their own slope bound \cite[Thm.~6.4]{BV22a} is of order $n^2$. \item[(K3)] The decomposition is known when $\dim S_k(\GL_2(A))=1$ \cite{BV23a}, when it is at most~$2$ \cite{DK23}, when every Hecke eigenvalue in level one has multiplicity less than $p$ \cite{dV26}, and, for $q$ \emph{odd}, when $\dim S_k(\Gamma_0(t))$ is even \cite{Dal26}. For $q=2$ and $k=2^m+1$ with $m\ge3$ none of these applies: $\dim S_k(\GL_2(A))=J_m\ge3$, the characteristic is even, and the level-one eigenvalues are multiple roots (Remark~\ref{rem:dv}). In characteristic $2$ the trace method of \cite{dV26} determines the eigenvalues of odd multiplicity and gives no information on those of even multiplicity. The recent strong Ramanujan bound of de Vries~\cite{dV26b} is a bound on traces; as stated there, it does not imply the corresponding bound for repeated Hecke eigenvalues in characteristic~$2$. \item[(K4)] For $q=2$ the slopes have been computed and tabulated: \cite[Ex.~6.10]{BV22a} for $k=9$, $17$, $33$, $65$, $129$ and \cite[Section~4]{Hat21} for $k\le23$. The elementary lower bound~\eqref{eq:elem} below is \cite[Lem.~2.3(3)]{Hat21}. \end{itemize} \subsection{Results} \begin{thmA}[\textsc{proved}] Let $S$ be the symmetric matrix over $\F_2$, indexed by the integers $p,q\ge0$ written in binary, with \[ S_{pq}=[p\AND q=0]+[p\sub q]+[q\sub p]\pmod 2 . \] Then $\det S[0{:}i]$ is nonzero if and only if $i$ is a Jacobsthal number, and $\cork S[0{:}i]$ is the distance from $i$ to the set $\{J_r\}$. Equivalently: let $L_N$ be the graph whose vertices are the integers $1,\dots,N$, viewed as nonempty finite sets of positive integers through their binary digits, two of them being adjacent when the sets are disjoint or nested. Then $L_N$ has an \emph{odd} number of perfect matchings if and only if $N+1$ is a Jacobsthal number. \end{thmA} \begin{thmC}[\textsc{proved}] For $k=2^m+1$, every slope of the Hecke operator $T_t$ on $S_k(\GL_2(A))$ is at most $J_m$. More precisely $\dim S_k(\GL_2(A))=J_m$, and $v_t(\det T_t)=J_m(J_m+1)/2$. \end{thmC} \begin{mainthm}[\textsc{proved}] For $q=2$ and $k=2^m+1$: \begin{itemize} \item[(i)] $S_k(\Gamma_0(t))=S_k^{\old}(\Gamma_0(t))\oplus S_k^{\new}(\Gamma_0(t))$, with $\dim S^{\old}=2J_m$ and $\dim S^{\new}=2J_{m-1}$; \item[(ii)] the characteristic polynomial of $U_t$ is $X^{J_m}\cdot P_m(X)\cdot(X^2+t^k)^{J_{m-1}}$, where $P_m$ is the characteristic polynomial of $T_t$ on $S_k(\GL_2(A))$; \item[(iii)] $k/2$ is the largest slope of $U_t$, it occurs with multiplicity exactly $2J_{m-1}$, carried by the newforms, and every nonzero slope of an oldform is at most $J_m