\documentclass{amsart} \usepackage{amssymb,amsmath} \usepackage{enumitem} \usepackage{xcolor} \usepackage{array} \usepackage{longtable} \usepackage{tikz} \usepackage{hyperref} \hypersetup{hidelinks, pdftitle={Improving Razmyslov's (1974) bound: the Nagata-Higman index is at most n\textasciicircum 2-2 for n>=3}, pdfauthor={Dave Goel}} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \newtheorem{conjecture}[theorem]{Conjecture} \newtheorem{mainthm}{Theorem} \renewcommand{\themainthm}{\Alph{mainthm}} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \newtheorem{problem}[theorem]{Problem} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newtheorem*{example}{Example} \newcommand{\sgn}{\operatorname{sgn}} \newcommand{\End}{\operatorname{End}} \newcommand{\Res}{\operatorname{Res}} \newcommand{\Stab}{\operatorname{Stab}} \newcommand{\Sym}{\operatorname{Sym}} \newcommand{\tr}{\operatorname{tr}} \newcommand{\Arc}{\operatorname{Arc}} \setlength{\emergencystretch}{3em} \makeatletter \def\@cite#1#2{{\@citestyle[\citeform{#1}\if@tempswa,~#2\fi]}}% \makeatother \title[Improving Razmyslov's bound]{Improving Razmyslov's (1974) bound:\\ the Nagata--Higman index is at most $n^2-2$ for $n\ge3$} \author{Dave Goel} \email{david44hello@gmail.com} \thanks{The Lean~4 proof of this paper's results:\newline \url{https://github.com/david44hello/papers_supporting_materials/tree/nagata-higman-index-v1.12/nagata-higman-index}.} \subjclass[2020]{Primary 16R10; Secondary 16N40, 16R30, 20C30} \keywords{Nagata--Higman theorem, nil algebras, nilpotency index, Kuzmin's conjecture, polynomial identities, symmetric group} \date{November 1, 2026} \begin{document} \begin{abstract} Let $d(n)$ be the least $m$ such that every associative (not necessarily unital) algebra over a field of characteristic zero satisfying the identity $x^n=0$ also satisfies $x_1x_2\cdots x_m=0$; this is the Nagata--Higman index. Razmyslov proved $d(n)\le n^2$ in 1974; Kuzmin proved $d(n)\ge n(n+1)/2$ and conjectured equality; this is the Kuzmin (1975) conjecture. We prove $d(n)\le n^2-2$ for $n\ge3$ (and, by a simpler case of the same method, $d(n)\le n^2-1$ for $n\ge2$); in particular $d(6)\le34$. To our knowledge, $n=6$ is the first case in which the new bound improves the known bounds, and this is the first improvement of Razmyslov's bound for general $n$. For $m=n^2-1$ and $m=n^2-2$, Formanek's criterion reduces $d(n)\le m$ to the statement that no nonzero linear combination of $(m+1)$-cycles lies in one particular isotypic block of the group algebra of $S_{m+1}$. For $m=n^2-1$ this statement follows from the branching rule; for $m=n^2-2$ we prove it using the Jucys--Murphy elements and a theorem of Frobenius on the characters of the alternating group. These bounds have also been proved in the Lean proof assistant. \end{abstract} \maketitle \section{Introduction}\label{sec:intro} Throughout, $F$ is a field of characteristic zero. An associative (not necessarily unital) $F$-algebra is \emph{nil of index $n$} if $x^n=0$ for each of its elements $x$. By the Nagata--Higman theorem, such an algebra is nilpotent: there is an $m$, depending only on $n$, such that $x_1x_2\cdots x_m=0$ for all elements $x_1,\dots,x_m$ of every such algebra. The least such $m$ is the \emph{Nagata--Higman index} $d(n)$. Razmyslov \cite{Ra74} proved $d(n)\le n^2$ in 1974; Kuzmin \cite{Ku75} proved $d(n)\ge n(n+1)/2$ in 1975 and conjectured that equality holds. For $n\ge3$ we lower Razmyslov's bound by two. The table and the list below give our main results, each with the best result known before this paper. \begin{table}[t] \small \renewcommand{\arraystretch}{1.1} \begin{center} \begin{tabular}{|>{\raggedright\arraybackslash}p{0.46\textwidth}|>{\raggedright\arraybackslash}p{0.43\textwidth}|} \hline \textbf{New result} & \textbf{Best result before this paper}\\ \hline $d(n)\le n^2-2$ for $n\ge3$, new for $n\ge6$ (Theorem~\ref{thm:B}) & $d(n)\le n^2$ (Razmyslov 1974 \cite{Ra74}), still called the best known upper bound in 2022 \cite{Ro22}\\ \hline $d(6)\le34$ & $d(6)\le36$ \cite{Ra74}\\ \hline invariants of $n\times n$ matrices: generated in degree \mbox{$\le n^2-2$} for $n\ge3$, new for $n\ge6$ (Corollary~\ref{cor:invariants}) & degree $\le n^2$ (\cite{Ra74} with \cite[Thm.~3.3]{Pr76}), ``at the moment the best result'' \cite[Thm.~5.14]{Pr26}\\ \hline \end{tabular} \end{center} \end{table} \begin{itemize}[leftmargin=*] \item \emph{Theorem~\ref{thm:A}: $d(n)\le n^2-1$ for every $n\ge2$.} This lowers Razmyslov's bound $n^2$ \cite{Ra74} by one; Theorem~\ref{thm:B} supersedes it for $n\ge3$. The proof (Section~\ref{sec:A}) uses the square partition $(n,\dots,n)$. \item \emph{Theorem~\ref{thm:B}: $d(n)\le n^2-2$ for every $n\ge3$.} This lowers Razmyslov's bound \cite{Ra74} by two; with Kuzmin's lower bound \cite{Ku75}, $n(n+1)/2\le d(n)\le n^2-2$. The proof (Section~\ref{sec:B}) uses the partition \mbox{$(n,\dots,n,n-1)$} and a theorem of Frobenius. \item \emph{$d(6)\le34$.} The values $d(n)$ for $n\le4$ are known \cite{Du35,Hi56,VL93}, and Higman's recursion with $d(4)=10$ gives $d(5)\le21$ \cite{Hi56}; so $n=6$ is the first case in which Theorem~\ref{thm:B} improves the known bounds. \item \emph{Corollary~\ref{cor:invariants}: for $n\ge3$, the invariants of $n\times n$ matrices are generated in degree at most $n^2-2$; this is new for $n\ge6$.} Before, the known bound for general $n$, uniform in the number of matrices, was $n^2$, from Razmyslov's bound and the theorem of Procesi and Formanek \cite[Thm.~3.3]{Pr76}, \cite[Thm.~6]{Fo86}; Procesi \cite[Thm.~5.14]{Pr26} still calls it ``at the moment the best result''. \end{itemize} \subsection{The Nagata--Higman theorem}\label{subsec:NH} The theorem is due to Dubnov and Ivanov \cite{DI43}, Nagata \cite{Na52} and Higman \cite{Hi56}, and is also called the Dubnov--Ivanov--Nagata--Higman theorem \cite[\S5]{Dr95}; Higman's proof gives $d(n)\le2^n-1$. The index $d(n)$ is also the least $m$ such that $x_1\cdots x_m$ lies in the T-ideal generated by $x^n$, that is, in the smallest ideal of the free non-unital algebra $F\langle x_1,x_2,\dots\rangle$ that contains $x^n$ and is stable under all endomorphisms (here $x$ is one of the letters); see \cite[\S1.3]{Ro22}. Indeed, the polynomials that vanish under every substitution of elements of an algebra $R$ form a T-ideal, and it contains $x^n$ if $R$ is nil of index $n$. Conversely, the quotient of $F\langle x_1,x_2,\dots\rangle$ by the T-ideal of $x^n$ is itself nil of index $n$: each of its elements is the image of a polynomial $h$ without constant term, and $h^n$ is the image of $x^n$ under the endomorphism that sends $x$ to $h$ and fixes the other letters. In words, a product vanishes in every algebra that is nil of index $n$ exactly when it is a consequence of the identity $x^n=0$. We use this form in Section~\ref{sec:criterion}. \subsection{Bounds and known values}\label{subsec:bounds} Razmyslov's proof \cite[final remark]{Ra74} uses the trace identities of the matrix algebra $M_n(F)$ and the Cayley--Hamilton theorem. By Procesi \cite[Thm.~3.3]{Pr76} and Formanek \cite[Thm.~6]{Fo86} (see also \cite{Fo87,Fo91}), the invariants of any number of $n\times n$ matrices under simultaneous conjugation are generated by traces of products of degree at most $d(n)$, and this degree bound is sharp when the number of matrices is large enough \cite[Thm.~5.6]{Dr95}. Kuzmin's lower bound \cite{Ku75} comes from the word $x_2x_1x_2x_1^2\cdots x_2x_1^{n-1}$: it has degree \mbox{$n(n+1)/2-1$} and does not lie in the T-ideal of $x^n$ (see \cite[Thm.~1.7]{Do18}), so neither does $x_1x_2\cdots x_{n(n+1)/2-1}$, which maps onto it under an endomorphism. The exact values $d(1)=1$, $d(2)=3$ and $d(3)=6$ are classical (Dubnov \cite{Du35}; see \cite[\S5]{Dr95} and, for $n=3$, \cite{Hi56}), and Vaughan-Lee \cite{VL93} computed $d(4)=10$ by computer. Higman's recursion $d(n)\le2d(n-1)+1$ \cite{Hi56} then gives $d(5)\le21$, and Kuzmin's bound gives $d(5)\ge15$. To our knowledge, in characteristic zero the bound $n^2$ had not been improved for general $n$ since 1974; see \cite[\S5]{Dr95}, \cite[\S1.3]{Ro22} and \cite[Thm.~5.14]{Pr26}. \subsection{Results} \begin{mainthm}\label{thm:A} For every $n\ge2$, $d(n)\le n^2-1$. \end{mainthm} \begin{mainthm}\label{thm:B} For every $n\ge3$, $d(n)\le n^2-2$. \end{mainthm} For $n=2$, Theorem~\ref{thm:A} only recovers the classical bound $d(2)\le3$. Theorem~\ref{thm:A} is the first, simpler case of the method: given Formanek's criterion (Proposition~\ref{prop:criterion}), its proof is short (Lemma~\ref{lem:onecorner}), but we have not found it in print. The gain over Razmyslov's bound is an additive constant; the gap to Kuzmin's lower bound is still of order $n^2/2$. By the theorem of Procesi and Formanek (Subsection~\ref{subsec:bounds}), for $n\ge6$ Theorem~\ref{thm:B} also lowers the known degree bound, uniform in the number of matrices, for the generators of the matrix invariants: \begin{corollary}\label{cor:invariants} For $n\ge3$ and every $k\ge1$, the algebra of polynomial invariants of $k$-tuples $(X_1,\dots,X_k)$ of $n\times n$ matrices over $F$ under simultaneous conjugation is generated by the traces $\tr(X_{i_1}\cdots X_{i_r})$ with $1\le r\le n^2-2$. \end{corollary} \begin{proof} First let $F$ be algebraically closed. By the first fundamental theorem of matrix invariants \cite[Thm.~5.3]{Dr95}, the invariants are generated by the traces $\tr(X_{i_1}\cdots X_{i_r})$; by the theorem of Procesi and Formanek (Subsection~\ref{subsec:bounds}), those with $r\le d(n)$ suffice; and $d(n)\le n^2-2$ by Theorem~\ref{thm:B}. Now let $F$ be arbitrary (of characteristic zero), with algebraic closure $\bar F$. Conjugation preserves degrees, so it suffices to treat invariants $g$ that are homogeneous of some degree $\delta$ in the $kn^2$ matrix entries. Writing $\gamma^{-1}=\det(\gamma)^{-1}\operatorname{adj}(\gamma)$ for an invertible matrix $\gamma$, invariance says $g(\gamma X_1\operatorname{adj}(\gamma),\dots,\gamma X_k\operatorname{adj}(\gamma))=\det(\gamma)^\delta g(X_1,\dots,X_k)$ for all invertible $\gamma$. Since $F$ is infinite, this holds for all invertible $\gamma$ over $F$ if and only if it holds as a polynomial identity in the entries of $\gamma$ and of the $X_i$ (the product of $\det(\gamma)$ and the difference of the two sides vanishes at every point, so it is the zero polynomial, and hence so is the difference); that is a system of linear equations, with integer coefficients, for the coefficients of $g$. So the invariants of degree $\delta$ over $F$ and over $\bar F$ are the solutions of the same linear system, and the two spaces have the same dimension. The products of total degree $\delta$ of traces $\tr(X_{i_1}\cdots X_{i_r})$ with $r\le n^2-2$ have integer coefficients and span the invariants of degree $\delta$ over $\bar F$; so their span over $F$ has the same dimension, and it is the whole space of invariants of degree $\delta$ over $F$. \end{proof} \subsection{Outline of the proofs}\label{subsec:outline} Let $m\ge n$ and $N=m+1$. Let $C_N$ be the span of the $N$-cycles in the group algebra $F[S_N]$. For a partition $\lambda\vdash N$, let $I_\lambda$ be the isotypic block of $\lambda$, that is, the minimal two-sided ideal of $F[S_N]$ that acts nontrivially on the irreducible $F[S_N]$-module $S^\lambda$ indexed by $\lambda$ (see (F2)); it acts as zero on the other irreducible modules. We identify $f\in F[S_N]$ with its coefficient function $\sigma\mapsto f(\sigma)$, and we say that $f$ \emph{vanishes on} a set of permutations if $f(\sigma)=0$ for every $\sigma$ in the set. Our starting point is Formanek's criterion \cite[Thm.~5]{Fo90}; related statements are in \cite[final remark]{Ra74}, \cite[Thm.~3.3]{Pr76} and \cite{St89}. We prove it in the following cycle form (Proposition~\ref{prop:criterion}): $d(n)\le m$ if and only if $C_N\cap B_n=0$, where $B_n$ is the sum of the blocks $I_\lambda$ over the partitions $\lambda\vdash N$ that fit in the $n\times n$ box, that is, have at most $n$ rows and at most $n$ columns. For example, for $n=2$ only $(2,1)$ fits when $N=3$. This is the nil analogue of the description of the multilinear trace identities of matrix algebras due to Helling, Razmyslov and Procesi \cite{He74,Ra74,Pr76}; see also \cite[Thm.~5.7]{Dr95}. The box has $n^2$ cells, so for $N>n^2$ no partition fits, and the criterion gives Razmyslov's bound (Remark~\ref{rem:razmyslov}). Theorems~\ref{thm:A} and~\ref{thm:B} treat the next two cases, $N=n^2$ and $N=n^2-1$; in each, exactly one partition fits (Figure~\ref{fig:shapes}). \begin{figure}[t] \centering \begin{tikzpicture}[x=0.75cm,y=0.75cm,font=\small] \fill[black!15] (2,-3) rectangle (3,-2); \draw[step=1] (0,-3) grid (3,0); \foreach \i in {1,2,3} \foreach \j in {1,2,3} {\pgfmathtruncatemacro{\c}{\j-\i}\node at (\j-0.5,0.5-\i) {$\c$};} \node at (1.5,-3.55) {$(3^3)\vdash 9$}; \begin{scope}[xshift=4.5cm] \fill[black!15] (2,-2) rectangle (3,-1); \fill[black!15] (1,-3) rectangle (2,-2); \draw[step=1] (0,-2) grid (3,0); \draw[step=1] (0,-3) grid (2,-2); \draw[dashed] (2,-3) rectangle (3,-2); \foreach \i in {1,2} \foreach \j in {1,2,3} {\pgfmathtruncatemacro{\c}{\j-\i}\node at (\j-0.5,0.5-\i) {$\c$};} \foreach \j in {1,2} {\pgfmathtruncatemacro{\c}{\j-3}\node at (\j-0.5,-2.5) {$\c$};} \node at (1.5,-3.55) {$\nu=(3^2,2)\vdash 8$}; \end{scope} \end{tikzpicture} \caption{The partitions that fit in the $n\times n$ box in the cases of Theorems~\ref{thm:A} and~\ref{thm:B}, drawn for $n=3$. Each cell $(i,j)$ shows its content $j-i$; the removable corners are shaded. Left: the square $(3^3)$, the case $N=n^2$; its only removable corner is the cell $(3,3)$. Right: $\nu=(3^2,2)$, the case $N=n^2-1$, which is the square with the cell $(3,3)$ (dashed) removed; its two removable corners, $(2,3)$ and $(3,2)$, have contents $+1$ and $-1$.}\label{fig:shapes} \end{figure} \begin{enumerate}[label=(\roman*),leftmargin=*, itemsep=1pt, topsep=2pt] \item The case $N=n^2$ (Theorem~\ref{thm:A}). Only the square $(n^n)=(n,\dots,n)$ fits. An element of $C_N$ vanishes on the subgroup $S_{N-1}$ of the permutations that fix $N$, since no $N$-cycle fixes a point. The square has only one removable corner (a cell whose removal leaves the diagram of a partition), so, by the branching rule, the restriction of $S^{(n^n)}$ to $S_{N-1}$ is irreducible. Hence an element of the block $I_{(n^n)}$ is determined by its coefficients on $S_{N-1}$, and an element of $C_N\cap I_{(n^n)}$ is zero (Lemma~\ref{lem:onecorner}). \item The case $N=n^2-1$ (Theorem~\ref{thm:B}). Only $\nu=(n^{n-1},n-1)$, the square with the cell $(n,n)$ removed, fits. Its two removable corners, $(n-1,n)$ and $(n,n-1)$, have contents $+1$ and $-1$ (the content of the cell $(i,j)$ is $j-i$). An element $f$ of $C_N$ vanishes on every permutation with a fixed point. Let $\rho_\nu$ be the representation of $F[S_N]$ on $S^\nu$, and let the \emph{star element} $X^{(j)}:=\sum_{i\ne j}(i\,j)$ be the sum of the $N-1$ transpositions that move $j$; thus $X^{(N)}=J_N$, where $J_k:=\sum_{i2$). On the group side, $C_3$ is spanned by the $3$-cycles $(1\,2\,3)$ and $(1\,3\,2)$, which are $\Phi(x_1x_2)$ and $\Phi(x_2x_1)$ in step~(1), and $B_2^{(3)}=I_{(2,1)}$, since only $(2,1)$ fits in the $2\times2$ box. Both $3$-cycles are even, so $f=a(1\,2\,3)+b(1\,3\,2)$ has the components \mbox{$(a+b)e_{(3)}$} and \mbox{$(a+b)e_{(1^3)}$} in the other two blocks; hence $C_3\cap I_{(2,1)}$ is spanned by $(1\,2\,3)-(1\,3\,2)$, and \mbox{$m!-\dim(C_3\cap B_2^{(3)})=2-1=\dim W_{2,2}$}, as Proposition~\ref{prop:criterion} says. In step~(7), this is the statement that the orthogonal complement of $\Phi(W_{2,2})=F\cdot\bigl((1\,2\,3)+(1\,3\,2)\bigr)$ in $C_3$ is $F\cdot\bigl((1\,2\,3)-(1\,3\,2)\bigr)$. \end{example} \begin{remark}[the direction used]\label{rem:forward} Theorems~\ref{thm:A} and~\ref{thm:B} use only the direction in which $C_N\cap B_n=0$ implies $d(n)\le m$. This direction needs only step~(7) and the inclusion $P(I)\subseteq\Phi(W_{n,m})$ of step~(6); the reverse inclusion gives the exact formula. \end{remark} \begin{remark}[Razmyslov's bound]\label{rem:razmyslov} Take $m=n^2$, so that $N=n^2+1$. No partition of $N$ fits in the $n\times n$ box, so $B_n^{(N)}=0$, and Proposition~\ref{prop:criterion} gives $d(n)\le n^2$. \end{remark} \section{Proof of Theorem~\ref{thm:A}}\label{sec:A} \begin{lemma}[the block of the square]\label{lem:onecorner} For $n\ge2$, $C_{n^2}\cap B_n^{(n^2)}=0$. \end{lemma} \begin{proof} Let $N=n^2$. The only partition of $n^2$ that fits in the $n\times n$ box is the square $\beta:=(n^n)$, so $B_n^{(N)}=I_\beta$. Let $f\in C_N\cap I_\beta$ and $M:=\rho_\beta(f)$. Then $f(\sigma)=(\dim S^\beta/N!)\tr(\rho_\beta(\sigma^{-1})M)$ by (F2). A permutation $\sigma$ in the stabilizer $S_{N-1}$ of the point $N$ is not an $N$-cycle, so $f(\sigma)=0$. The square has only one removable corner, so by (F5) $\Res_{S_{N-1}}S^\beta$ is the module $S^\nu$ with $\nu=(n^{n-1},n-1)$, which is absolutely irreducible by (F2); by (F6), the $\rho_\beta(\sigma)$ with $\sigma\in S_{N-1}$ span $\End(S^\beta)$. Hence $\tr(ZM)=0$ for every $Z\in\End(S^\beta)$, so $M=0$ and $f=0$. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:A}] Lemma~\ref{lem:onecorner} and Proposition~\ref{prop:criterion} with $m=n^2-1\ge n$ give $d(n)\le n^2-1$. \end{proof} \section{Proof of Theorem~\ref{thm:B}}\label{sec:B} We follow the outline in Subsection~\ref{subsec:outline}(ii). \subsection{Reduction to one block}\label{subsec:reduction} Let $N=n^2-1$. A partition of $N$ that fits in the $n\times n$ box leaves out exactly one cell of the box, and this cell must be $(n,n)$, the only cell of the box whose removal leaves a diagram. Thus the only partition of $N$ that fits is $\nu:=(n^{n-1},n-1)$, the box with the cell $(n,n)$ removed. Hence $B_n=I_\nu$, and by Proposition~\ref{prop:criterion} it suffices to show $C_N\cap I_\nu=0$. By (F1), $d(n)$ does not depend on $F$, so from now on we take $F=\mathbb{C}$, which \ref{T3} below requires. For example, for $n=3$, $\nu=(3,3,2)$ and $N=8$; the contents of its cells are $0,1,2$ in the first row, $-1,0,1$ in the second and $-2,-1$ in the third, and its two removable corners are the cells $(2,3)$ and $(3,2)$, of contents $+1$ and $-1$. For $f\in I_\nu$ put $M:=\rho_\nu(f)$; then, by (F2), \[ f(\sigma)=\frac{\dim S^\nu}{N!}\tr\bigl(\rho_\nu(\sigma^{-1})M\bigr).\tag{$*$} \] \subsection{Facts from the representation theory of \texorpdfstring{$S_N$}{S\_N}}\label{subsec:facts} Besides (F2) and (F5)--(F7), we use three facts. Below, $\rho$ is $\rho_\lambda$ for a partition $\lambda\vdash N$ (from Subsection~\ref{subsec:stars} on, $\lambda=\nu$). A standard tableau $T$ of shape $\lambda$ is a filling of the diagram of $\lambda$ with $1,\dots,N$, increasing along the rows and down the columns; for an entry $k$, $c_T(k)$ is the content of the cell of $k$ in $T$, and $T_{\le k}$ is the shape formed by the cells of $1,\dots,k$. The \emph{Young basis} $\{v_T\}$ of $S^\lambda$ is defined, up to nonzero scalars, by following $T$ down the chain $S_N\supset S_{N-1}\supset\dots\supset S_1$ \cite{OV96}, \cite[\S3.2]{CST10}: by the branching rule (F5), $S^\lambda$ contains exactly one $F[S_{N-1}]$-submodule isomorphic to $S^{T_{\le N-1}}$, this submodule contains exactly one $F[S_{N-2}]$-submodule isomorphic to $S^{T_{\le N-2}}$, and so on; the last of these submodules is a line, and $v_T$ spans it. (This is not Young's natural basis of polytabloids \cite{Ja78}.) \begin{enumerate}[label=(T\arabic*),leftmargin=*] \item\label{T1} \emph{Jucys--Murphy elements.} Put $J_k:=\sum_{ic$; as $\lambda=\lambda'$, the hook of $(i,i)$ has $h_{ii}=2(\lambda_i-i)+1$ cells. These numbers are odd, and they are distinct because $\lambda_i-i$ decreases strictly with $i$. Their sum is $N$, since every cell $(a,b)$ of $\lambda$ lies in the hook of exactly one diagonal cell, namely $(c,c)$ with $c=\min(a,b)$ (a cell of $\lambda$, since $c\le a$ and $c\le b$). So $h(\lambda):=(h_{11},\dots,h_{rr})$ is a cycle type of $S_N$. \emph{Frobenius' theorem} \cite{Fr01}, \cite[Thm.~2.5.13]{JK81} says that $\chi_+(\sigma)=\chi_-(\sigma)$ unless $\sigma$ has cycle type $h(\lambda)$. So $\psi$ vanishes on every permutation whose cycle type is not $h(\lambda)$. \end{enumerate} \begin{example} Let $\lambda=(2,1)$, so that $N=3$, and let $T_{12}$ and $T_{13}$ be the standard tableaux with first rows $1,2$ and $1,3$. In the basis $v_{T_{12}},v_{T_{13}}$, \ref{T1} gives \mbox{$\rho((1\,2))=\rho(J_2)=\operatorname{diag}(1,-1)$}, and \ref{T2} gives the diagonal entries $1/(-1-1)=-\frac12$ and $1/(1+1)=\frac12$ of $\rho((2\,3))$. With the $v_T$ scaled as in Young's orthogonal form \cite[\S3.4.2]{CST10}, \[ \rho((1\,2))=\begin{pmatrix}1&0\\0&-1\end{pmatrix},\qquad \rho((2\,3))=\begin{pmatrix}-\frac12&\frac{\sqrt3}2\\[2pt]\frac{\sqrt3}2&\frac12\end{pmatrix}. \] An operator that anticommutes with both matrices has zero diagonal entries (by the first) and opposite off-diagonal entries (by the second), so it is a multiple of $\Gamma_0:=\left(\begin{smallmatrix}0&1\\-1&0\end{smallmatrix}\right)$. As $\Gamma_0^2=-\mathrm{id}$, the operator $\Gamma$ of \ref{T3} is $\pm i\,\Gamma_0$; this is why \ref{T3} needs $\mathbb C$. For $\Gamma=i\,\Gamma_0$, $\psi$ vanishes on the identity and on the transpositions, and $\psi((1\,2\,3))=-i\sqrt3= -\psi((1\,3\,2))$, as Frobenius' theorem says for $h(\lambda)=(3)$. \end{example} \subsection{Star elements and permutations with a fixed point}\label{subsec:stars} For $j\in[N]$ define the \emph{star element} $X^{(j)}:=\sum_{i\ne j}(i\,j)\in F[S_N]$, the sum of the $N-1$ transpositions that move $j$, and put $Y_j:=\rho(X^{(j)})$. Note that $X^{(N)}=J_N$ and $gX^{(j)}g^{-1}=X^{(g(j))}$ for $g\in S_N$. The shape $\nu$ has exactly two removable corners, $a_+:=(n-1,n)$ of content $+1$ and $a_-:=(n,n-1)$ of content $-1$; the other row ends $(i,n)$, $i2$ by Proposition~\ref{prop:criterion}; with Theorem~\ref{thm:A}, $d(2)=3$. \end{remark} \section*{Acknowledgments} The author thanks Claudio Procesi for helpful suggestions. This work was done overwhelmingly by AI models and supervised by the author. The models were Anthropic's Claude and OpenAI's ChatGPT. The author takes full responsibility for the paper. \begin{thebibliography}{CST10} \bibitem[CST10]{CST10} T.~Ceccherini-Silberstein, F.~Scarabotti and F.~Tolli, \emph{Representation Theory of the Symmetric Groups: The Okounkov--Vershik Approach, Character Formulas, and Partition Algebras}, Cambridge Stud. Adv. Math. \textbf{121}, Cambridge Univ. Press, 2010. \bibitem[Do18]{Do18} M.~Domokos, Polynomial bound for the nilpotency index of finitely generated nil algebras, \emph{Algebra Number Theory} \textbf{12} (2018), 1233--1242. \bibitem[Dr95]{Dr95} V.~Drensky, Commutative and noncommutative invariant theory, in: \emph{Math. and Education in Math.} (Proc. 24th Spring Conf. Union Bulg. Math., Svishtov, 1995), 14--50; arXiv:2302.10052. \bibitem[Du35]{Du35} J.~Dubnov, Sur une g\'en\'eralisation de l'\'equation de Hamilton--Cayley et sur les invariants simultan\'es de plusieurs affineurs, \emph{Proc. Seminar Vector Tensor Analysis, Moscow State Univ.} \textbf{2/3} (1935), 351--367. \bibitem[DI43]{DI43} Ya.~S. Dubnov and V.~K. Ivanov, Sur l'abaissement du degr\'e des polyn\^omes en affineurs, \emph{C. R. (Doklady) Acad. Sci. URSS} \textbf{41} (1943), 95--98. \bibitem[Et11]{Et11} P.~Etingof, O.~Golberg, S.~Hensel, T.~Liu, A.~Schwendner, D.~Vaintrob and E.~Yudovina, \emph{Introduction to Representation Theory}, Stud. Math. Libr. \textbf{59}, Amer. Math. Soc., 2011; arXiv:0901.0827 (theorem numbers as in arXiv version~5). \bibitem[Fo86]{Fo86} E.~Formanek, Generating the ring of matrix invariants, in: \emph{Ring Theory (Antwerp, 1985)}, Lecture Notes in Math. \textbf{1197}, Springer, 1986, 73--82. \bibitem[Fo87]{Fo87} E.~Formanek, The invariants of $n\times n$ matrices, in: \emph{Invariant Theory}, Lecture Notes in Math. \textbf{1278}, Springer, 1987, 18--43. \bibitem[Fo90]{Fo90} E.~Formanek, The Nagata--Higman theorem, \emph{Acta Appl. Math.} \textbf{21} (1990), 185--192. \bibitem[Fo91]{Fo91} E.~Formanek, \emph{The Polynomial Identities and Invariants of $n\times n$ Matrices}, CBMS Regional Conf. Ser. Math. \textbf{78}, Amer. Math. Soc., 1991. \bibitem[Fr01]{Fr01} G.~Frobenius, \"Uber die Charaktere der alternirenden Gruppe, \emph{Sitzungsber. K\"onigl. Preu\ss. Akad. Wiss. Berlin} (1901), 303--315. \bibitem[Go26]{FV} D.~Goel, Improving Razmyslov's (1974) bound: the Nagata--Higman index is at most $n^2-2$ for $n\ge3$, and other miscellanea, Zenodo, 2026, \url{https://doi.org/10.5281/zenodo.23624549}. \bibitem[He74]{He74} H.~Helling, Eine Kennzeichnung von Charakteren auf Gruppen und assoziativen Algebren, \emph{Comm. Algebra} \textbf{1} (1974), 491--501. \bibitem[Hi56]{Hi56} G.~Higman, On a conjecture of Nagata, \emph{Proc. Cambridge Philos. Soc.} \textbf{52} (1956), 1--4. \bibitem[Ja78]{Ja78} G.~D. James, \emph{The Representation Theory of the Symmetric Groups}, Lecture Notes in Math. \textbf{682}, Springer, 1978. \bibitem[JK81]{JK81} G.~D. James and A.~Kerber, \emph{The Representation Theory of the Symmetric Group}, Encyclopedia Math. Appl. \textbf{16}, Addison-Wesley, 1981. \bibitem[Ju74]{Ju74} A.-A.~A. Jucys, Symmetric polynomials and the center of the symmetric group ring, \emph{Rep. Math. Phys.} \textbf{5} (1974), 107--112. \bibitem[Ku75]{Ku75} E.~N. Kuz'min, On the Nagata--Higman theorem (Russian), in: \emph{Mathematical Structures, Computational Mathematics, Mathematical Modelling}, Sofia, 1975, 101--107. \bibitem[Mu81]{Mu81} G.~E. Murphy, A new construction of Young's seminormal representation of the symmetric groups, \emph{J. Algebra} \textbf{69} (1981), 287--297. \bibitem[Na52]{Na52} M.~Nagata, On the nilpotency of nil-algebras, \emph{J. Math. Soc. Japan} \textbf{4} (1952), 296--301. \bibitem[OV96]{OV96} A.~Okounkov and A.~Vershik, A new approach to representation theory of symmetric groups, \emph{Selecta Math. (N.S.)} \textbf{2} (1996), 581--605. \bibitem[Pr76]{Pr76} C.~Procesi, The invariant theory of $n\times n$ matrices, \emph{Adv. Math.} \textbf{19} (1976), 306--381. \bibitem[Pr26]{Pr26} C.~Procesi, Fermionic matrices and super Cayley--Hamilton algebras, arXiv:2605.22745 (2026). \bibitem[Ra74]{Ra74} Yu.~P. Razmyslov, Trace identities of full matrix algebras over a field of characteristic zero, \emph{Izv. Akad. Nauk SSSR Ser. Mat.} \textbf{38} (1974), 723--756; English transl.: \emph{Math. USSR-Izv.} \textbf{8} (1974), 727--760. \bibitem[Ro22]{Ro22} A.~Romano, On the T-ideal generated by the identity $f=x^n$, arXiv:2212.05994 (2022). \bibitem[St89]{St89} E.~Strickland, On algebras satisfying the identity $x^N=0$, \emph{Comm. Algebra} \textbf{17} (1989), 2795--2799. \bibitem[VL93]{VL93} M.~Vaughan-Lee, An algorithm for computing graded algebras, \emph{J. Symbolic Comput.} \textbf{16} (1993), 345--354. \end{thebibliography} \end{document}