Gradings, graded identities, $*$-identities and graded $*$-identities of an algebra of upper triangular matrices
Extracted source text · arXiv 2411.06964
\documentclass{amsart}
\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage[english]{babel}
\usepackage{amsmath}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{graphicx}
\usepackage[colorlinks=true, allcolors=blue]{hyperref}
\usepackage{xcolor}
\usepackage{amsthm}
\newtheorem{theorem}{Theorem}
\newtheorem{lemma}{Lemma}
\newtheorem{proposition}{Proposition}
\newtheorem{corollary}{Corollary}
\theoremstyle{definition}
\newtheorem{definition}{Definition}
\theoremstyle{remark}
\newtheorem{remark}{Remark}
\DeclareMathOperator{\Id}{Id}
\DeclareMathOperator{\Span}{Span}
\DeclareMathOperator{\sign}{sign}
\title[Identities of an algebra of upper triangular matrices]{Gradings, graded identities, $*$-identities and graded $*$-identities of an algebra of upper triangular matrices}
\author[J. A. Gomez Parada]{Jonatan Andres Gomez Parada}
\address{Department of Mathematics \\ IMECC, UNICAMP \\
Sérgio Buarque de Holanda, 651, 13083-859 Campinas, SP, Brazil}
\email{j211980@dac.unicamp.br}
\thanks{This study was financed in part by the Coordena\c c\~ao de Aperfei\c coamento de Pessoal de N\'{\i}vel Superior - Brasil (CAPES) -
Finance Code 001.}
\author[P. Koshlukov]{Plamen Koshlukov}
\address{Department of Mathematics \\ IMECC, UNICAMP \\
Sérgio Buarque de Holanda, 651, 13083-859 Campinas, SP, Brazil}
\email{plamen@unicamp.br}
\thanks{P. Koshlukov was partially supported by FAPESP Grant 2018/23690-6 and by CNPq Grant 307184/2023-4}
\date{}
\begin{document}
\begin{abstract}
Let $K \langle X\rangle$ be the free associative algebra freely generated over the field $K$ by the countable set $X = \{x_1, x_2, \ldots\}$. If $A$ is an associative $K$-algebra, we say that a polynomial $f(x_1,\ldots, x_n) \in K \langle X\rangle$ is a polynomial identity, or simply an identity in $A$ if $f(a_1,\ldots, a_n) = 0$ for every $a_1$, \dots, $a_n \in A$.
Consider $\mathcal{A}$ the subalgebra of $UT_3(K)$ given by:
\[ \mathcal{A} = K(e_{1,1} + e_{3,3}) \oplus Ke_{2,2} \oplus Ke_{2,3} \oplus Ke_{3,2} \oplus Ke_{1,3} , \]
where $e_{i,j}$ denote the matrix units. We investigate the gradings on the algebra $\mathcal{A}$, determined by an abelian group, and prove that these gradings are elementary. Furthermore, we compute a basis for the $\mathbb{Z}_2$-graded identities of $\mathcal{A}$, and also for the $\mathbb{Z}_2$-graded identities with graded involution. Moreover, we describe the cocharacters of this algebra.
\end{abstract}
\keywords{Polynomial identities; graded identities; identities with involution; upper triangular matrices}
\subjclass{16R10, 16W10, 16R50, 16W50}
\maketitle
\section*{Introduction}
Let $K$ be a field and let $K\langle X \rangle$ be the free associative algebra freely generated by the countable set of indeterminates $X$. One can view $K\langle X\rangle$ as the set of all polynomials in the non-commuting variables from the set $X$. Given an associative algebra $A$, a polynomial $f$ in $K\langle X\rangle$ is called a {\sl polynomial identity} of $A$ if $f$ evaluates to zero when its variables are substituted with arbitrary elements of $A$. An algebra satisfying a non-zero polynomial identity is called a {\sl PI-algebra}. The set of all identities for $A$ is denoted by $T(A)$. Clearly $T(A)$ is an ideal in $K\langle X\rangle$. Moreover it is closed under endomorphisms of the free algebra $K\langle X\rangle$. It can be seen that every such ideal coincides with $T(A)$ for some $A$.
Among the algebras that satisfy non-zero polynomial identities, those that have been of greatest interest in the development of the theory of polynomial identities and that will play a significant role in this paper include the Grassmann algebra, the full matrix algebras $M_n(K)$, and $UT_n(K)$, the algebra of the upper triangular matrices of order $n$.
One initial problem to be considered in the theory of algebras with polynomial identities, which we will address here, is determining the set of all identities satisfied by a particular algebra, as well as a generating set for them. In the case of a PI-algebra $A$ over a field of characteristic zero, it is well known that the polynomial identities of $A$ follow from the multilinear ones. Therefore, we can restrict our study to multilinear polynomials.
In the particular case of the algebra $UT_n(K)$, its polynomial identities are well known. If $K$ is any infinite field, then a basis for the polynomials identities of $UT_n(K)$ is given by the polynomial $[x_1, x_2] \cdots [x_{2k-1}, x_{2k}]$. Considering the graded case, A. Valenti in \cite{valenti2002graded} studied the graded identities and the graded cocharacters of $UT_2(K)$. In \cite{di2004} the authors classified the elementary gradings on the upper triangular matrix algebras, and described their graded identities. In \cite{valenti2007group} Valenti and Zaicev proved that every group grading on the algebra $UT_n(K)$ is isomorphic to an elementary grading. Later on Cirrito considered in \cite{cirrito2013gradings} the graded identities of $UT_3(K)$.
Involutions on the algebra $UT_n(K)$ have also been extensively studied. An {\sl involution} on an algebra $A$ is an antiautomorphism of order two, that is, a linear map $* \colon A \to A$ satisfying $(ab)^{*} = b^{*}a^{*}$ and $(a^{*})^{*} = a$, for every $a$, $b \in A$. In the classification of involutions, such ones are called involutions of the first kind. The involutions on $UT_n(K)$ were described in \cite{di2006involutions}. In that paper, the authors also described the identities with involution for the algebras $UT_2(K)$ and $UT_3(K)$. Graded involutions on $UT_n(K)$ were described by Valenti and Zaicev in \cite{valenti2009graded}.
We are interested in a particular subalgebra of $UT_3(K)$. Consider $\mathcal{A}$ the subalgebra of $UT_3(K)$ given by
\[ \mathcal{A} = K(e_{1,1} + e_{3,3}) \oplus Ke_{2,2} \oplus Ke_{1,2} \oplus Ke_{2,3} \oplus Ke_{1,3} , \]
where $e_{i,j}$ denote the matrix units, that is, $A$ consists of the matrices $\begin{pmatrix} d & a & c \\ 0 & g & b \\ 0 & 0 & d \end{pmatrix}$.
The polynomial identities of $\mathcal{A}$ were described by Gordienko in \cite{gordienko2009regev}. Furthermore, in \cite{pktcm}, \cite{koshlukov2013polynomial} the authors considered the generic algebra of $M_{1,1}(E)$ in two generators. It was shown in that paper that its polynomial identities are the same as the ones of $\mathcal{A}$. We are interested in the identities of the algebra $\mathcal{A}$ when considering additional structures (grading, involution, etc.), with the aim of determining a new PI-equivalence like in the case of traditional polynomial identities. Here we point out that the research in \cite{koshlukov2013polynomial} was motivated by a question posed by A. Berele, about the centre of the generic algebra of $M_{1,1}(E)$ in $d\ge 2$ generators. It is well known that the algebra generated by $d$ generic matrices for $M_n(K)$ is a (noncommutative) domain. Hence its centre is a commutative domain, and it can me embedded into its field of fractions. Several important questions in PI theory could be solved by using this simple trick.
The algebra $\mathcal{A}$, with involution, was also considered in \cite{MISHCHENKO200066}.
We draw the readers' attention that in characteristic 0, the polynomial identities of $M_{1,1}(E)$ were determined by Popov \cite{popov}, and they follow from the identities $[[x_1,x_2]^2, x_1]$ and $[[x_1,x_2],[x_3,x_4], x_5]$. If one considers the generic algebra in two generators for $M_{1,1}(E)$ (or the algebra $\mathcal{A}$), the identities follow from the following three: $[[x_1,x_2][x_3,x_4],x_5]$, $[x_1,x_2][x_3,x_4][x_5,x_6]$, and from the standard identity $s_4$ of degree 4. Neither of these is an identity for $M_{1,1}(E)$, though.
In this paper, we first study the gradings on the algebra $\mathcal{A}$, when the grading group is abelian. Additionally, we determine a basis for the $\mathbb{Z}_2$-graded identities of $\mathcal{A}$, and also for the identities with involution, and for the $\mathbb{Z}_2$-graded identities with graded involution. We also determine the corresponding cocharacter sequence.
\section{Preliminaries}
\subsection{Graded identities}
Let $K$ be a fixed field of characteristic zero and $G$ a group. All algebras we consider will be associative and over $K$.
The algebra $A$ is a $G$-graded algebra if $ A = \oplus_{g\in G} A_g $ is a direct sum of the vector subspaces $A_g$, satisfying the relations
$A_gA_h \subseteq A_{gh}$ for every $ g$, $h\in G$. The subspaces $A_g$ are called the {\sl homogeneous components of $A$}, and the elements of each $A_g$ are called the {\sl homogeneous elements of $A$} of homogeneous degree $g$. We say that an associative algebra is a {\sl superalgebra} if $A$ is a $\mathbb{Z}_2$-graded algebra.
Let $G$ be a group, and consider the set of indeterminates $X = \cup_{g\in G} X_g$, where $X_g = \{x_1^{(g)}, x_2^{(g)}, \dots \}$ is a countable infinite set for each $g$. We say that the indeterminates in $X_g$ have a homogeneous degree $g$, and the homogeneous degree of a monomial $x_{i_1}^{(g_{j_1})} \cdots x_{i_m}^{(g_{j_m})} \in K\langle X \rangle$ is given by $g_{j_1} \cdots g_{j_m}$. We will write $K\langle X \rangle^{gr}$ to denote the graded algebra $K\langle X \rangle$ with $X = \cup_{g\in G} X_g$.
\begin{definition}
Let $f(x_{i_1}^{(g_1)},x_{i_2}^{(g_2)}, \dots, x_{i_r}^{(g_r)}) \in K\langle X \rangle^{gr}$ be a polynomial. If $A = \oplus_{g\in G}A_g$ is a $G$-graded algebra then $f$ is a {\sl $G$-graded polynomial identity} (or simply a {\sl $G$-graded identity}) for $A$ if $f(a_{i_1}^{(g_1)} , a_{i_2}^{(g_2)} , \dots , a_{i_r}^{(g_r)} ) = 0$ in $A$ for every homogeneous substitution $a^{(g_t)} \in A_{g_t}$. We denote by $\Id_G(A)$ or $T_G(A)$ the ideal of all graded identities of $A$ in $K\langle X \rangle^{gr}$.
\end{definition}
The ideal $\Id_G(A)$ is closed under all $G$-graded endomorphisms of $K\langle X \rangle^{gr}$; such ideals are called {\sl $G$-graded T-ideals}.
We are going to consider $G = \mathbb{Z}_2$. In this case we write $x_i^{(0)} = y_i$ and $x_i^{(1)} = z_i$, and also $K\langle X \rangle^{gr}$ as $K\langle Y,Z \rangle $. It is known that in characteristic zero, every graded polynomial identity is equivalent to a finite system of multilinear graded identities. Thus, given the free associative $\mathbb{Z}_2$-graded algebra $K \langle Y,Z \rangle$ we consider $P_{n_1,n_2}$ the space of multilinear polynomials in $n_1$ variables of degree $0$ and $n_2$ variables of degree $1$.
For $n_1$, $n_2 \geq 0$, consider the action of the group $S_{n_1}\times S_{n_2} $ on $P_{n_1,n_2}$ given by
\[
(\omega,\tau)f(y_{1},\dots,y_{n_1},z_{1},\dots,z_{n_2})
= f(y_{\omega(1)},\dots,y_{\omega(n_1)},z_{\tau(1)},\dots,z_{\tau(n_2)}),
\]
where $(\omega, \tau) \in S_{n_1} \times S_{n_2} $ and $f \in P_{n_1,n_2}$. Then $P_{n_1,n_2}$ is a left $S_{n_1}\times S_{n_2}$-module.
Given a $\mathbb{Z}_2$-graded algebra $A$, denote by $P_{n_1,n_2}(A)$ the quotient space \[ P_{n_1,n_2}(A) = \frac{ P_{n_1,n_2}}{ P_{n_1,n_2}\cap \Id_{\mathbb{Z}_2}(A)}.\]
Since $T_2$-ideals are invariant under permutations of the variables, it follows that $P_{n_1,n_2} \cap \Id_{\mathbb{Z}_2}(A)$ is an $S_{n_1} \times S_{n_2} $-submodule of $P_{n_1,n_2}$. Thus, $P_{n_1,n_2}(A)$ has a structure of an $S_{n_1} \times S_{n_2} $-module, and its character $\chi_{n_1,n_2}(A)$ is called the {\sl $(n_1,n_2)$-th graded cocharacter of $A$}.
The $S_{n_1} \times S_{n_2}$-characters are obtained from the outer tensor product of irreducible characters of $S_{n_1}$ and $S_{n_2}$, and we have 1--1 correspondence between the irreducible characters of $S_{n_1} \times S_{n_2}$ and the pair of partitions $(\omega, \tau)$, where $\omega \vdash n_1$ and $\tau \vdash n_2$. We denote by $\chi_{\omega} \otimes \chi_{\tau}$ the irreducible $S_{n_1} \times S_{n_2}$-character associated to the pair of partitions $(\omega, \tau)$. Thus,
\begin{equation}
\chi_{n_1,n_2}(A) = \sum_{(\omega,\tau)\vdash (n_1,n_2)} m_{\omega,\tau} \chi_{\omega}\otimes \chi_{\tau},
\end{equation}
where $m_{\omega,\tau}$ is the multiplicity of $\chi_{\omega}\otimes \chi_{\tau}$. The multiplicity $m_{\omega,\tau}$ is equal to the number of linearly independent highest weight vectors corresponding to a standard Young Tableau of shape $(\omega, \tau)$, modulo $\Id_{\mathbb{Z}_2}(A)$ (see \cite{drensky2000free} for more details).
\subsection{Identities with involution}
The following type of polynomials identities that will be consider are the identities of an algebra endowed with an involution. To this end, let $X = \{x_1, x_2, \ldots \}$ be a countable set of non-commuting variables and consider $K\langle X, *\rangle = K\langle x_1, x_1^*, x_2, x_2^*, \ldots \rangle$, the {\sl free associative algebra with involution} in $X$ over $K$. By defining $y_i = x_i + x_i^*$ and $z_i = x_i - x_i^*$ for each $i = 1, 2, \ldots$, we consider $K\langle X, *\rangle = K\langle Y, Z \rangle = K\langle y_1, z_1, y_2, z_2, \ldots \rangle$ as generated by symmetric and skew-symmetric variables. The elements of $K\langle X, *\rangle$ will be called $*$-polynomials.
Note that we use $K\langle Y, Z \rangle$ to denote both the free $\mathbb{Z}_2$-graded algebra and the free algebra with involution. Therefore, the structure of the algebra $A$ will determine which free associative algebra is being considered.
\begin{definition}
A $*$-polynomial $f\left( y_1,\dots,y_n,z_1,\dots,z_m \right) \in K\langle Y \cup Z \rangle$ is called a {\sl $*$-polynomial identity} of an algebra with involution $(A,*)$ if \[ f\left( u_1,\dots,u_n,v_1,\dots,v_m \right) = 0 \text{ for all } u_i \in A^+ \text{ and } v_j \in A^- . \]
\end{definition}
Given an algebra with involution $(A,*)$ we denote by $\Id^*(A)$ or $T^*(A)$ the set of $*$-polynomial identities of $A$.
In a similar way to the case of $\mathbb{Z}_2$-graded identities, we consider the vector space of multilinear polynomials in symmetric and skew-symmetric variables, that is, we consider $P_{n_1,n_2}$ the space of multilinear polynomials in $n_1$ symmetric variables and $n_2$ skew-symmetric variables. Given $A$ an algebra with involution, we denote by $P_{n_1,n_2}(A)$ the quotient space \[ P_{n_1,n_2}(A) = \frac{ P_{n_1,n_2}}{ P_{n_1,n_2}\cap \Id^*(A)}.\]
Denote by $\Gamma$ the unitary subalgebra of $K\langle Y, Z\rangle$ generated by the elements from $Z$ and all non-trivial (long) commutators in the free variables of $K\langle Y, Z\rangle$. The elements of $\Gamma$ are called {\sl $Y$-proper polynomials}. We denote by $\Gamma_{n_1,n_2}$ the vector space
\[ \Gamma_{n_1,n_2} = \Gamma \cap P_{n_1,n_2}. \]
Given $A$ a unitary algebra with involution, we consider the spaces \[ \Gamma_{n_1,n_2}(A) = \frac{ \Gamma_{n_1,n_2}}{ \Gamma_{n_1,n_2}\cap \Id^*(A)}. \]
Since $K$ is of characteristic zero, $\Id^*(A)$is generated, as a $T^*$-ideal, by its multilinear $Y$-proper polynomials.
\subsection{Graded involutions}
\begin{definition}
An involution $*$ on a $G$-graded algebra $A = \oplus_{g \in G} A_g$ is said to be a {\sl graded involution} if $A_g^*= A_g$ for all $g \in G$.
\end{definition}
Similarly to the case of the free associative algebra with involution, consider the free associative graded algebra with graded involution freely generated by homogeneous symmetric and homogeneous skew-symmetric variables.
Consider the free $\mathbb{Z}_2$-algebra with involution $K\langle Y \cup Z, * \rangle$, generated by symmetric and skew elements of even and odd degree, that is \[ K\langle Y \cup Z, * \rangle = K \langle y_{1}^{+}, y_{1}^{-}, z_{1}^{+}, z_{1}^{-}, y_{2}^{+}, y_{2}^{-}, z_{2}^{+}, z_{2}^{-}, \dots \rangle \]
where $y_{i}^{+}$ stands for a symmetric variable of even degree, $y_{i}^{-}$ for a skew variable of even degree, $z_{i}^{+}$ for a
symmetric variable of odd degree and $z_{i}^{-}$ for a skew variable of odd degree.
Given $A$ a $\mathbb{Z}_2$-graded algebra with graded involution and a graded $*$-polynomial $f(y_{1}^{+},\dots,y_{m}^{+},y_{1}^{-},\dots,y_{n}^{-},z_{1}^{+},\dots,z_{s}^{+},z_{1}^{-},\dots,z_{t}^{-}) \in F\langle Y \cup Z, * \rangle$, then $f$ is a {\sl graded $*$-polynomial identity} for $A$, if for all $u_{1}^{+},\dots,u_{m}^{+} \in A_{0}^{+}$, $u_{1}^{-},\dots,u_{n}^{-} \in A_{0}^{-}$, $v_{1}^{+},\dots,v_{s}^{+} \in A_{1}^{+}$, and $v_{1}^{-},\dots,v_{t}^{-} \in A_{1}^{-}$,
\[ f(u_{1}^{+},\dots,u_{m}^{+}, u_{1}^{-},\dots,u_{n}^{-}, v_{1}^{+},\dots,v_{s}^{+},v_{1}^{-},\dots,v_{t}^{-}) = 0. \]
We denote by $\Id_{\mathbb{Z}_2}^{*}(A) = \{ f \in K\langle Y \cup Z\rangle \mid f \equiv 0 \text{ on } A \}$ the $T_2^{*}$-ideal of graded $*$-identities of $A$, i.e., $\Id_{\mathbb{Z}_2}^{*}(A)$ is an ideal of $K\langle Y \cup Z\rangle$, invariant under all $\mathbb{Z}_2$-graded endomorphisms of the free $\mathbb{Z}_2$-graded algebra commuting with the involution $*$. Again, in characteristic zero, every graded $*$-identity is equivalent to a finite system of multilinear graded $*$-identities. So, consider $P_{n_1,n_2,n_3,n_4}$ the space of multilinear polynomials in $n_1$ symmetric variables of degree $0$, $n_2$ skew-symmetric variables of degree $0$, $n_3$ symmetric variables of degree $1$ and $n_4$ skew-symmetric variables of degree $1$.
We consider the cocharacters of $\mathbb{Z}_2$-graded algebras with graded involutions. For $n_1$, $n_2$, $n_3$, $n_4 \geq 0$, consider the action of the group $S_{n_1}\times S_{n_2} \times S_{n_3} \times S_{n_4}$ on $P_{n_1,n_2, n_3, n_4}$ given by
\[ \begin{split}
(\omega,\sigma, \tau, \rho)f(y_{1}^{+},\dots,y_{n_1}^{+},y_{1}^{-},\dots,y_{n_2}^{-},z_{1}^{+},\dots,z_{n_3}^{+},z_{1}^{-},\dots,z_{n_4}^{-}) \\
= f(y_{\omega(1)}^{+},\dots,y_{\omega(n_1)}^{+},y_{\sigma(1)}^{-},\dots,y_{\sigma(n_2)}^{-},z_{\tau(1)}^{+},\dots,z_{\tau(n_3)}^{+},z_{\rho(1)}^{-},\dots,z_{\rho(n_4)}^{-}),
\end{split} \]
where $(\omega,\sigma, \tau, \rho) \in S_{n_1} \times S_{n_2} \times S_{n_3} \times S_{n_4}$ and $f \in P_{n_1,n_2, n_3, n_4}$.
Given a $\mathbb{Z}_2$-graded algebra $A$ with a graded involution, denote by $P_{n_1,n_2,n_3,n_4}(A)$ the quotient space \[ P_{n_1,n_2,n_3,n_4}(A) = \frac{ P_{n_1,n_2,n_3,n_4}}{ P_{n_1,n_2,n_3,n_4}\cap \Id_{\mathbb{Z}_2}^*(A)}.\]
We examine the action of $S_{n_1} \times S_{n_2} \times S_{n_3} \times S_{n_4}$ on $P_{n_1,n_2,n_3,n_4}(A)$. Since $T_2^*$-ideals are invariant under permutations of the variables, then $P_{n_1,n_2,n_3,n_4} \cap \Id_{\mathbb{Z}_2}^*(A)$ is a left $S_{n_1} \times S_{n_2} \times S_{n_3} \times S_{n_4}$-submodule of $P_{n_1,n_2,n_3,n_4}$. Thus, $P_{n_1,n_2,n_3,n_4}(A)$ has a structure of left $S_{n_1} \times S_{n_2} \times S_{n_3} \times S_{n_4}$-module, and its character $\chi_{n_1,n_2,n_3, n_4}(A)$ is called the {\sl $(n_1,n_2,n_3,n_4)$-th cocharacter of $A$}.
The $S_{n_1} \times S_{n_2} \times S_{n_3} \times S_{n_4}$-characters are obtained from the outer tensor product of irreducible characters of $S_{n_1}$, $S_{n_2}$, $S_{n_3}$ and $S_{n_4}$, and we have 1--1 correspondence between the irreducible characters of $S_{n_1} \times S_{n_2} \times S_{n_3} \times S_{n_4}$ and the 4-tuples of partitions $(\omega,\sigma, \tau, \rho)$, where $\omega \vdash n_1 $, $\sigma \vdash n_2$, $\tau \vdash n_3$ and $\rho \vdash n_4$. We denote by $\chi_{\omega}\otimes \chi_{\sigma} \otimes \chi_{\tau}\otimes \chi_{\rho}$ the irreducible $S_{n_1} \times S_{n_2} \times S_{n_3} \times S_{n_4}$-character associated to the 4-tuple of partitions $(\omega,\sigma, \tau, \rho)$. Thus,
\begin{equation}
\chi_{n_1,n_2,n_3, n_4}(A) = \sum_{(\omega,\sigma,\tau,\rho)\vdash (n_1,n_2,n_3,n_4)} m_{\omega,\sigma,\tau,\rho} \chi_{\omega}\otimes \chi_{\sigma} \otimes \chi_{\tau}\otimes \chi_{\rho},
\end{equation}
where $m_{\omega,\sigma,\tau,\rho}$ is the multiplicity of $\chi_{\omega}\otimes \chi_{\sigma} \otimes \chi_{\tau}\otimes \chi_{\rho}$. The multiplicity $m_{\omega,\sigma,\tau,\rho}$ is equal to the number of linearly independent highest weight vectors corresponding to a standard Young Tableau of shape $(\omega,\sigma, \tau, \rho)$, modulo $\Id_{\mathbb{Z}_2}^*(A)$.
\section{Gradings on $A$}
We fix the algebra $ A = K(e_{1,1} + e_{3,3}) \oplus Ke_{2,2} \oplus Ke_{1,2} \oplus Ke_{2,3} \oplus Ke_{1,3} $.
We want to characterize the gradings on $A$ reducing them to elementary ones. As the main result of this section, we have that the gradings on the algebra $A$ are equivalent to elementary gradings.
\begin{definition}
Given a triple $\hat{g} = (g_1,g_2,g_3)$ of elements of an arbitrary group $G$ and $1\leq i,j \leq 3$, we set the degrees for the unitary matrices as $\deg e_{i,j} = g^{-1}_{i} g_{j}$ for $i \neq j$ or $i=j=2$ , and $\deg (e_{1,1} + e_{3,3}) = 1_G$. Given $g \in G$ let $A_g = \Span\{e_{i,j} \mid g^{-1}_{i}g_{j}= g\}$ for $g \neq 1_G$ and $A_{1_G} = \Span\left( \{e_{i,j} \mid g^{-1}_{i}g_{j}= 1_G\} \cup \{ e_{1,1} + e_{3,3} \} \right)$. Then $A = \oplus_{g\in G} A_g$ is a $G$-grading of $A$ called the {\sl elementary} $G$-grading defined by the $3$-tuple $\hat{g}$.
\end{definition}
We start considering the idempotents of the algebra $A$.
\begin{proposition}
\label{idempotent}
If $e \in A$ is an idempotent element, then $e$ is conjugated to a diagonal element of $A$.
\end{proposition}
\begin{proof}
If $e = \begin{pmatrix} x & a & c \\ 0 & y & b \\ 0 & 0 & x \end{pmatrix} \neq \mathbf{0} $ and $e^2 = e$ then $ \begin{pmatrix} x^2 & a(x+y) & 2xc + ab \\ 0 & y^2 & b(x+y) \\ 0 & 0 & x^2 \end{pmatrix} = \begin{pmatrix} x & a & c \\ 0 & y & b \\ 0 & 0 & x \end{pmatrix}$ which implies that $x,y \in \{0,1\}$ where $x$ and $y$ are not simultaneously zero. So, we consider the following cases:
\begin{itemize}
\item If $x = 0$, $y=1$ then
$e = \begin{pmatrix} 0 & a & ab \\ 0 & 1 & b \\ 0 & 0 & 0 \end{pmatrix}$. If $q = \begin{pmatrix} 1 & -a & 0 \\ 0 & 1 & b \\ 0 & 0 & 1 \end{pmatrix}$ a simple computation shows that $qeq^{-1} = \begin{pmatrix} 0 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 0 \end{pmatrix}$.
\item If $x = 1$, $y=0$ then
$e = \begin{pmatrix} 1 & a & -ab \\ 0 & 0 & b \\ 0 & 0 & 1 \end{pmatrix}$. Choose $q = \begin{pmatrix} 1 & a & 0 \\ 0 & 1 & -b \\ 0 & 0 & 1 \end{pmatrix}$, then $qeq^{-1} = \begin{pmatrix} 1 & 0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 1 \end{pmatrix}$.
\item If $x = 1$, $y=1$ then
$e = \begin{pmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{pmatrix}$.
\end{itemize}
\end{proof}
\begin{remark} \label{rm:conjg}
Note that if we have two orthogonal idempotent elements, then each one of these is conjugated to an element of the set $\{ (e_{1,1} + e_{3,3}), e_{2,2} \}$.
\end{remark}
\begin{remark}
Let $A = \oplus_{g \in G} A_g$ be graded by an abelian group. Since for $g$, $h \in G$, $[A_g , A_h] \subset A_{gh} + A_{hg} \subset A_{gh}$, it follows that the commutator subalgebra $[A, A]$ is a nilpotent non-zero graded ideal of A.
\end{remark}
\begin{lemma} \label{lm:orthInden}
Let $A = \oplus_{g \in G} A_g$ be graded by an abelian group $G$ with identity element $1_G \in G$. Then $A_{1_G}$ contains $2$ orthogonal idempotents.
\end{lemma}
\begin{proof}
Let $E$ be the identity element of $A$. Since $E \in A_{1_G}$, there exists a non trivial maximal semisimple subalgebra $B$ of $A_{1_G}$.
Let $C$ be one of the simple summands of $B$ and let $e$ be its unit element.
By Proposition \ref{idempotent} it follows that $e$ is conjugated to a diagonal idempotent. Hence either $e$ and $E - e$ are two orthogonal idempotents or $e = E$ and $C = B = \Span{E}$.
Consider the case $e = E$ and $B = \Span{E}$, we shall show that this leads to a contradiction and hence is impossible.
First, note that any homogeneous element of $A$ is either nilpotent or invertible.
In fact, suppose that $a \in A_g$ is not nilpotent. For $m$ large enough the elements $a$, $a^2$,\dots, $a^m$ are linearly dependent and homogeneous. It follows that $g$ must have finite order $k$ and $a^k \in A_{1_G}$. Moreover being not nilpotent the element $a^k$ does not lie in the Jacobson radical $J(A_{1_G})$ of $A_{1_G}$. Since $A_{1_G}/J(A_{1_G}) \cong K\cdot E$, it follows that $a^k - \lambda E$ is nilpotent for some $\lambda \in K$, $\lambda \neq 0$. This means that $a^k$ and hence also $a$ is invertible.
Next we prove that the Jacobson radical $J(A)$ of $A$, i.e. the subalgebra of all
strictly upper-triangular matrices does not contain non-zero homogeneous elements.
$J(A)$ is homogeneous in the $G$-grading since $\varphi(J(A)) = J(A)$ for any $\varphi \in {\rm Aut }A$.
In fact suppose by contradiction that $0 \neq a \in A_g$ is nilpotent and consider its left annihilator
$L_a = \{x \in A \mid xa = 0\}$ and right annihilator $R_a = \{x \in A \mid ax = 0\}$, these are graded subspaces of $A$. Then, as the elements of $L_a$ and $R_a$ are zero divisors they are not invertible, hence they are nilpotent. By our hypothesis $a$ is nilpotent, $a = \begin{pmatrix} 0 & a_1 & a_3 \\ 0 & 0 & a_2 \\ 0 & 0 & 0 \end{pmatrix}$ and $a^2 = \begin{pmatrix} 0 & 0 & a_1 a_2 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{pmatrix}$. Note that, if $a^2 \neq 0$, $a^2 \in A_{g^2}$ and then $e_{2,2} \in L_{a^2}$, a contradiction. Now, if $a^2 = 0$ then $a = \begin{pmatrix} 0 & a_1 & a_3 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \end{pmatrix}$ or $a = \begin{pmatrix} 0 & 0 & a_3 \\ 0 & 0 & a_2 \\ 0 & 0 & 0 \end{pmatrix}$, so $e_{2,2} \in L_a$ or $e_{2,2} \in R_a$, once again a contradiction. Therefore, $J(A)$ has no nonzero homogeneous elements. But this gives us a contradiction because $[A, A] \subset J(A)$.
Thus, we have $e$ and $E - e$ are two orthogonal idempotents belonging to $A_{1_G}$.
\end{proof}
\begin{lemma} \label{lm:homog}
Let $A = \oplus_{g \in G} A_g$ be $G$-graded. Then the grading is elementary if and only if all matrix units $e_{i,j}$, $1\leq i < j \leq 3$, $e_{2,2}$ and $(e_{1,1} + e_{3,3})$ are homogeneous.
\end{lemma}
\begin{proof}
If the $G$-grading is elementary then those matrices are homogeneous by definition. Suppose that the said matrices are homogeneous. If we set $g_1= 1_G$, $g_2 = \deg e_{1,2}$ and $g_3 = g_2\deg e_{2,3}$ then the triple $(g_1,g_2,g_3)$ satisfies the conditions for the grading to be elementary.
\end{proof}
\begin{lemma} \label{lm:lm}
Let $A = \oplus_{g \in G} A_g$ be $G$-graded. Then the grading is elementary if and only if the matrices $e_{2,2}$ and $(e_{1,1} + e_{3,3})$ belong to $A_{1_G}$.
\end{lemma}
\begin{proof}
It is clear that if the grading is elementary then $e_{2,2}$, $(e_{1,1} + e_{3,3}) \in A_{1_G}$.
Now, assume that $e_{2,2}$, $(e_{1,1} + e_{3,3}) \in A_{1_G}$ and note that
\begin{itemize}
\item $e_{1,2} \in (e_{1,1} + e_{3,3}) A e_{2,2}$ with $\dim (e_{1,1} + e_{3,3}) A e_{2,2}= 1$,
\item $e_{2,3} \in e_{2,2} A (e_{1,1} + e_{3,3})$ with $\dim e_{2,2} A (e_{1,1} + e_{3,3})= 1$,
\item $e_{1,3} \in (e_{1,1} + e_{3,3}) A (e_{1,1} + e_{3,3}) = K(e_{1,1} + e_{3,3}) \oplus K e_{1,3}$.
\end{itemize}
Since $(e_{1,1} + e_{3,3}) A e_{2,2}$, $(e_{1,1} + e_{3,3}) A (e_{1,1} + e_{3,3})$ and $e_{2,2} A (e_{1,1} + e_{3,3})$ are graded subalgebras then we can conclude that the elements $e_{1,2}$, $e_{1,3}$ and $e_{1,2}$ are homogeneous. Thus, by Lemma \ref{lm:homog} the grading is elementary.
\end{proof}
\begin{lemma} \label{lm:idempDiago}
Let $A = \oplus_{g \in G} A_g$ be $G$-graded. Then there exist two orthogonal idempotents that are simultaneously diagonalizable and belonging to $A_{1_G}$.
\end{lemma}
\begin{proof}
Suppose $A$ is $G$-graded, then the identity matrix $E$ is homogeneous.
Also $J$, the Jacobson radical of $A$ is homogeneous. So is $J^2=\Span(e_{13})$. By the Wedderburn-Malcev theorem, $A\cong K+K+J$, a direct sum of vector spaces, where $K+K$ is a subalgebra. Hence we have two orthogonal idempotents. We show we can choose these homogeneous and simultaneously diagonalizable.
First take the intersection of the left and right annihilators of $J^2$, it is homogeneous. But this is exactly the span $\Span(e_{1,2},e_{1,3},e_{2,3},e_{2,2})$. There exists an idempotent in it, moreover all idempotents in it are of the form $\begin{pmatrix}0&a&ac\\ 0&1&c\\ 0&0&0 \end{pmatrix}$.
Suppose the element above is the homogeneous idempotent, call it $t$.
Now, considering $A$ acting on a $3$-dimensional vector space with a basis $\{e_1, e_2, e_3\}$, we obtain
\[
te_1=0, \quad te_2=ae_1+e_2, \quad te_3=ace_1+ce_2.
\]
Now consider the basis $f_1$, $f_2$, $f_3$ such that $tf_1=0$, $tf_2=f_2$, $tf_3=0$. Such a basis can be obtained as
\[
f_1=e_1, \quad f_2=ae_1+e_2, \quad f_3= -ce_2+e_3.
\]
($f_1$ and $f_3$ are a basis of the kernel of $t$, and $f_2$ of its image.)
The change of basis (matrix) is $P=\begin{pmatrix} 1&a&0\\ 0&1&-c\\
0&0&1\end{pmatrix}$.
Form $I-t$, this is an idempotent which is orthogonal to $t$, and clearly the same $P$ diagonalizes it. As $t$ is homogeneous and $I$ also is, then $I-t$ is homogeneous.
\end{proof}
\begin{theorem}\label{th:gradingofA}
Let $G$ be an abelian group and $K$ a field. Suppose that the $K$-algebra $A = \oplus_{g \in G} A_g$ is $G$-graded. Then $A$, as a $G$-graded algebra, is isomorphic to $A$ with an elementary $G$-grading.
\end{theorem}
\begin{proof}
Let $A = \oplus_{g \in G} A_g$ be $G$-graded. By Lemma \ref{lm:idempDiago} the homogeneous component $A_{1_G}$ contains two orthogonal idempotents simultaneously diagonalizable by a element $P$. So as a $G$-graded algebra $A$ is isomorphic to $A' = \oplus_{g\in G} A'_g$ where $A'_g = P^{-1}A_g P$. Note that in $A'$ the matrix $e_{2,2}$ and the matrix identity $E$ lie in $A'_1$. Therefore, by Lemma \ref{lm:lm}, $A'$ has an elementary $G$-grading.
\end{proof}
Next, we will consider the graded identities of the algebra $A$, where the grading is given by the group $\mathbb{Z}_2$.
Due to Theorem \ref{th:gradingofA}, we have $3$ different non-trivial $\mathbb{Z}_2$-gradings on $A$. Thus, for $A= A_0 \oplus A_1$ we have the possibilities:
\begin{equation}\label{eq:gra1}
A_0 = \left\{
\begin{pmatrix}
d & 0 & c \\
0 & g & 0 \\
0 & 0 & d
\end{pmatrix} \mid d,g,c \in K \right\}, \qquad
A_1 = \left\{ \begin{pmatrix}
0 & a & 0 \\
0 & 0 & b \\
0 & 0 & 0
\end{pmatrix} \mid a,b \in K \right\},
\end{equation}
or
\begin{equation}\label{eq:gra2}
A_0 = \left\{
\begin{pmatrix}
d & a & 0 \\
0 & g & 0 \\
0 & 0 & d
\end{pmatrix} \mid d,g,a \in K \right\}, \qquad
A_1 = \left\{ \begin{pmatrix}
0 & 0 & c \\
0 & 0 & b \\
0 & 0 & 0
\end{pmatrix} \mid c, b \in K \right\},
\end{equation}
or
\begin{equation}\label{eq:gra3}
A_0 = \left\{
\begin{pmatrix}
d & 0 & 0 \\
0 & g & b \\
0 & 0 & d
\end{pmatrix} \mid d,g,b \in K \right\}, \qquad
A_1 = \left\{ \begin{pmatrix}
0 & a & c \\
0 & 0 & 0 \\
0 & 0 & 0
\end{pmatrix} \mid a,c \in K \right\}.
\end{equation}
Denote by $\mathcal{A}^1$ the graded algebra $A$ with grading given by (\ref{eq:gra1}), $\mathcal{A}^2$ the graded algebra $A$ with grading given by (\ref{eq:gra2}), and $\mathcal{A}^3$ the graded algebra $A$ with grading given by (\ref{eq:gra3}).
\section{Graded polynomial identities of $\mathcal{A}^1$}
It is easy to see that the followings are graded identities of $\mathcal{A}^1$.
\begin{itemize}
\item[(a)] $[y_1, y_2] = 0$,
\item[(b)] $z_1z_2z_3 = 0$,
\item[(c)] $[y_1,z_1z_2] = 0$.
\end{itemize}
The identity (c) is a direct consequence of identity (a).
Note that $P_{m,n}(\mathcal{A}^1) = 0$ if $n\geq 3$. So, we consider the cases when $n=0,1,2$. In the case $n=0$, by the identity (a) it is clear that
\begin{equation} \label{eq:A_1n=0}
P_{m,0}(\mathcal{A}^1) = \Span\{y_1y_2\cdots y_m\}.
\end{equation}
\subsection{Case $n=1$}
Given a monomial $\mu$ in $P_{m,1}(\mathcal{A}^1)$, from (a) we can reorder the variables $y_i$ such that modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$ \begin{equation} \label{eq:A_1n=1}
\mu = y_{i_1}y_{i_2}\cdots y_{i_s}z_1 y_{j_1}y_{j_2}\cdots y_{j_t}
\end{equation} where $ i_1 < i_2 \cdots < i_s$, $ j_1 < j_2 \cdots < j_t$ and $s+t = m$, $s,t \geq 0$.
\begin{proposition}\label{pro:A_1n=1}
The monomials of the form (\ref{eq:A_1n=1}) are linearly independent modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$.
\end{proposition}
\begin{proof}
Let $f$ be a sum of monomials of the form (\ref{eq:A_1n=1}) \[ f = \sum_{I,J}\alpha_{I,J}y_{i_1}y_{i_2}\cdots y_{i_s}z_1 y_{j_1}y_{j_2}\cdots y_{j_t}.\]
Note that if $y_{i_k} = \beta_{i_k}(e_{1,1} + e_{3,3})$, $z_1 = \lambda e_{1,2}$ and $y_{j_l} = \eta_{j_l}e_{2,2}$ then \[ f = \sum_{I,J}\alpha_{I,J}\beta_{i_1}\beta_{i_2}\cdots \beta_{i_s}\lambda \eta_{j_1}\eta_{j_2}\cdots \eta_{j_t} \cdot e_{1,2}.\]
Suppose that $f$ is a polynomial identity of $\mathcal{A}^1$ and that there exist $\alpha_{I_0,J_0} \neq 0$. Consider the evaluation $y_{i_k} = e_{1,1} + e_{3,3}$ for $i_{k} \in I_0$, $y_{j_k} = e_{2,2}$ for $j_{k} \in J_0$ and $z_1= e_{1,2}$, then $\alpha_{I_0,J_0} = 0$, a contradiction. So, the monomials of the form (\ref{eq:A_1n=1}) are linearly independent.
\end{proof}
\subsection{Case $n=2$}
Given a monomial $\mu$ in $P_{m,2}(\mathcal{A}^1)$, again from (a) we can reorder the variables $y_i$ such that modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$ \begin{equation*}
\mu = y_{i_1}y_{i_2}\cdots y_{i_s}z_{l_1} y_{j_1}y_{j_2}\cdots y_{j_t} z_{l_2} y_{h_1}y_{h_2}\cdots y_{h_r}
\end{equation*} where $ i_1 < i_2 \cdots < i_s$, $ j_1 < j_2 \cdots < j_t$, $ h_1 < h_2 \cdots < h_r$ and $s+t+r = m$, $s,t,r \geq 0$.
As a consequence of (c) we have $z_{l_1}y_{j}z_{l_2}y_h = y_h z_{l_1}y_{j}z_{l_2} $. Therefore, $P_{m,2}(\mathcal{A}^1)$ is spanned by monomials of the form
\begin{equation} \label{eq:A_1n=2}
y_{i_1}y_{i_2}\cdots y_{i_s}z_{l_1} y_{j_1}y_{j_2}\cdots y_{j_t} z_{l_2}
\end{equation}
where $ i_1 < i_2 \cdots < i_s$, $ j_1 < j_2 \cdots < j_t$, and $s+t = m$, $s,t \geq 0$.
\begin{proposition} \label{pro:A_1n=2}
The monomials of the form (\ref{eq:A_1n=2}) are linearly independent modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$.
\end{proposition}
\begin{proof}
Note that if $ y_{h_k} = \begin{pmatrix} a_{h_k} & 0 & c_{h_k} \\
0 & b_{h_k} & 0 \\
0 & 0 & a_{h_k}
\end{pmatrix} $ and $ z_{l_i} = \begin{pmatrix} 0 & d_{l_i} & 0 \\
0 & 0 & e_{l_i} \\
0 & 0 & 0
\end{pmatrix} $ then \[ y_{i_1}y_{i_2}\cdots y_{i_s}z_{l_1} y_{j_1}y_{i_2}\cdots y_{j_t} z_{l_2} = a_{i_1}\cdots a_{i_s} d_{l_1} b_{j_1}\cdots b_{j_t}d_{l_2} \cdot e_{1,3}. \]
Let $f$ be a sum of monomials of the form (\ref{eq:A_1n=2})
\begin{align*} f &= \sum_{I,J,L}\alpha_{I,J,L}y_{i_1}y_{i_2}\cdots y_{i_s}z_{l_1} y_{j_1}y_{j_2}\cdots y_{j_t}z_{l_2}\\
& = \sum_{I,J,L}\alpha_{I,J,L}a_{i_1}\cdots a_{i_s} d_{l_1} b_{j_1}\cdots b_{j_t}e_{l_2} \cdot e_{1,3}.
\end{align*}
Suppose $f$ is a polynomial identity of $\mathcal{A}^1$ and there exists $\alpha_{I_0,J_0,L_0} \neq 0$. Consider the evaluation $y_{i_k} = e_{1,1} + e_{3,3}$ for $i_{k} \in I_0$, $y_{j_k} = e_{2,2}$ for $j_{k} \in J_0$, $z_{l_1}= e_{1,2}$, and $z_{l_2}= e_{2,3}$. Then, $\alpha_{I_0,J_0,L_0} = 0$, a contradiction. So, the monomials in the form (\ref{eq:A_1n=2}) are linearly independent.
\end{proof}
As a consequence of (\ref{eq:A_1n=0}), Proposition \ref{pro:A_1n=1} and Proposition \ref{pro:A_1n=2}, we have the following result.
\begin{theorem} \label{th:A_1}
The identities (a) and (b), form a basis for the $\mathbb{Z}_2$-graded identities for the algebra $\mathcal{A}^1$.
\end{theorem}
\section{Graded polynomial identities of $\mathcal{A}^2$}
Modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^2)$ we have that \begin{equation} \label{eq:m1-m4}
z_{j_1}z_{j_2} = 0 \text{ and } [y_{i_1}, y_{i_2}][y_{i_3}, y_{i_4}] = 0
\end{equation}
Once again, we consider the space $P_{m,n}$ of multilinear polynomials in the variables $y_{i_1}, y_{i_2}, \dots, y_{i_m}$, $z_{j_1}, z_{j_2}, \dots, z_{j_n}$. Since $ z_{j_1}z_{j_2} = 0$, we have $P_{m,n} = 0 $ if $n\geq 2$. Thus we consider the cases $n=0$ and $n=1$.
\subsection{Case $n=0$ }
By (\ref{eq:m1-m4}) we have that $ [ y_{i_1}, y_{i_2}][y_{i_3}, y_{i_4}] = 0$, then considering proper polynomials, we have only sums of commutators. Also from (\ref{eq:m1-m4}) and since \[ [[y_{i_1}, y_{i_2}],[y_{i_3}, y_{i_4}]] = [y_{i_1}, y_{i_2}, y_{i_3}, y_{i_4}] - [y_{i_1}, y_{i_2}, y_{i_4}, y_{i_3}], \] given a commutator we can reorder the variables in the way $[ y_{k}, y_{l} , y_{i_1}, y_{i_2},\dots, y_{i_{m-2}}]$ where $i_1 < i_2 < \cdots < i_{m-2}$, and by Jacobi identity we can only consider the case $[ y_{i_1}, y_{i_2},\dots, y_{i_{m}} ]$ where $i_1 > i_2 < i_3 < \cdots < i_{m-2}$.
Also, note that the subalgebra $\mathcal{A}^2_0$ is isomorphic to the algebra $UT_2(K)$. So, the polynomial $ [ y_{i_1}, y_{i_2}][y_{i_3}, y_{i_4}]$ is a basis of the identities in the variables $y_{i}$.
\subsection{Case $n=1$}
In this case, we consider once again proper polynomials, that is, sums of products of left normed commutators.
Note that since $ [y_{i_1},y_{i_2}] [y_{i_3},y_{i_4}] = 0$ and we have only one $z_j$ then we can consider sums of commutators with products of two commutators.
Now, since $ [y_{i_1},y_{i_2}] = \nu e_{1,2}$, then \begin{equation}\label{eq:N2}
z_{j_1}[y_{i_1},y_{i_2}] = 0.
\end{equation}
hence in the case when we have products of two commutators we get that $z_{j_1}$ lies in the second commutator.
By direct computation, we also have the following identity:
\begin{equation} \label{eq:N3}
[[y_{i_1},y_{i_2}]z_{j_1}, y_{i_3}] = 0,
\end{equation}
Note that from (\ref{eq:N2}) $ [y_{i_1},y_{i_2},z_{j_1}] = [y_{i_1},y_{i_2}]z_{j_1}$ and using the identity (\ref{eq:N3}) we have $ [y_{i_1},y_{i_2},z_{j_1},y_{i_3}] = 0$. Therefore, if $z_{j_1}$ lies in a commutator $\omega$, that commutator has the form $\omega = [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_k}] $ or $\omega = [y_{i_1},y_{i_2}, \dots,y_{i_k},z_{j_1}] = [y_{i_1},y_{i_2}, \dots,y_{i_k}]z_{j_1} $.
Let us consider the case $\omega = [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_k}] $. Note that by Jacobi identity $ [z_{j_1},y_{i_1},y_{i_2}] + [y_{i_1},y_{i_2},z_{j_1}] + [y_{i_2},z_{j_1},y_{i_1}] = 0 $, then
$ [z_{j_1},y_{i_2},y_{i_1}] = [z_{j_1},y_{i_1},y_{i_2}] + [y_{i_1},y_{i_2}]z_{j_1}$. So we can write the variables $y_i$ in any order in $\omega$.
If $\omega = [y_{i_1},y_{i_2}, \dots,y_{i_k},z_{j_1}] = [y_{i_1},y_{i_2}, \dots,y_{i_k}]z_{j_1} $, as in the case $P_{m,0}$, we can reorder the variables $y_{i_1},y_{i_2}, \dots,y_{i_k}$ such that $[y_{i_1},y_{i_2}, \dots,y_{i_k}]z_{j_1}$ is a linear combination of polynomials $[y_{i_l},y_{i_1},y_{i_2}, \dots ,\widehat{y_{i_l}}, \dots,y_{i_s}]z_{j_1}$ where $i_{l} > i_1 < i_2 < \cdots < i_k$.
At this point, we have that if $\mathcal{B}$ is the space of proper polynomials in the variables $z_{j_1}$, $y_{i_1}$, $y_{i_2}$, \dots, $y_{i_k}$, modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^2)$, $\mathcal{B}$ is spanned by \begin{itemize}
\item $ [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_k}]$, $i_1 < i_2 < \cdots < i_k$,
\item $ [y_{i_l},y_{i_1},y_{i_2}, \dots ,\widehat{y_{i_l}}, \dots,y_{i_k}]z_{j_1}$, $i_{l} > i_1 < i_2 < \cdots < i_k$,
\item $ [y_{i_l},y_{i_1},y_{i_2}, \dots ,\widehat{y_{i_l}}, \dots,y_{i_s}][z_{j_1}, y_{h_1},y_{h_2},\dots,y_{h_t} ]$, $i_{l} > i_1 < i_2 < \cdots < i_k$, $h_1 < h_2 < \cdots < h_s$.
\end{itemize}
\begin{proposition} \label{pr:A_2}
The polynomial \begin{equation}
[y_{i_1},y_{i_2}][z_{j_1},y_{i_3}] + [y_{i_1},y_{i_2},y_{i_3}]z_{j_1}
\end{equation}
is a polynomial identity of the algebra $\mathcal{A}^2$.
\end{proposition}
\begin{proof}
Note that \begin{align*}
[y_{i_1},y_{i_2}][z_{j_1},y_{i_3}] & = [y_{i_1},y_{i_2}]z_{j_1}y_{i_3} - [y_{i_1},y_{i_2}]y_{i_3}z_{j_1} \stackrel{(\ref{eq:N3})}{=} y_{i_3}[y_{i_1},y_{i_2}]z_{j_1} - [y_{i_1},y_{i_2}]y_{i_3}z_{j_1} \\
& = [y_{i_3},[y_{i_1},y_{i_2}]]z_{j_1} = -[y_{i_1},y_{i_2},y_{i_3}]z_{j_1}.\qedhere
\end{align*}
\end{proof}
Hence from Proposition \ref{pr:A_2} we consider linear combinations of
\begin{itemize}
\item $ [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_k}]$, $i_1 < i_2 < \cdots < i_k$,
\item $ [y_{i_l},y_{i_1},y_{i_2}, \dots ,\widehat{y_{i_l}}, \dots,y_{i_k}]z_{j_1}$, $i_{l} > i_1 < i_2 < \cdots < i_k$.
\end{itemize}
Note that, if
$ y_{i_k} = \begin{pmatrix}
d_{i_k} & a_{i_k} & 0 \\
0 & g_{i_k} & 0 \\
0 & 0 & d_{i_k}
\end{pmatrix} $
and
$ z_{j_1} = \begin{pmatrix} 0 & 0 & c_{j_1} \\
0 & 0 & b_{j_1} \\
0 & 0 & 0
\end{pmatrix} $, then
\[
[z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_m}] = (-1)^{m+1}b_{j_1} \prod_{k=1}^{m-1} (d_{i_k} - g_{i_k})a_{i_m} \cdot e_{1,3} + b_{j_1} \prod_{k=1}^{m} (d_{i_k} - g_{i_k}) \cdot e_{2,3}
\]
and
\[
[y_{i_l},y_{i_1},y_{i_2}, \dots ,\widehat{y_{i_l}}, \dots,y_{i_k}]z_{j_1}
= \{
\left[ a_{i_1}(d_{i_l} - g_{i_l}) + a_{i_l}(g_{i_l} - d_{i_l}) \right] \prod_{k \neq l}^{m} (g_{i_k} - d_{i_k})
\} b_{j_1} \cdot e_{2,3}
\]
\begin{theorem}\label{th:A_2}
The following identities form a basis for the $\mathbb{Z}_2$-graded identities of the algebra $\mathcal{A}^2$:
\begin{itemize}
\item[$(i)$] $ z_{i_1}z_{i_2}$,
\item[$(ii)$] $ [y_{i_1},y_{i_2}][y_{i_3},y_{i_4}] $,
\item[$(iii)$] $ z_{i_1}[y_{i_1},y_{i_2}] $,
\item[$(iv)$] $ [[y_{i_1},y_{i_2}]z_{j_1},y_{i_3}]$.
\end{itemize}
\end{theorem}
\begin{proof}
We want to prove that the polynomials $ [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_m}]$ with $i_1 < i_2 < \cdots < i_m$, and $ [y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}]z_{j_1} $ with $h_{l} > h_1 < h_2 < \cdots < h_m$ are linearly independent modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^2)$. Consider $f$ the polynomial given by \[ f = \alpha[z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_m}] + \sum_{l=2}^{m}\beta_l [y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}]z_{j_1} \] and suppose that $f\in \Id_{\mathbb{Z}_2}(\mathcal{A}^2)$. Now, considering the evaluation $z_{j_1} = e_{2,3}$ and $y_{i} = (e_{1,1} + e_{3,3})$ for $i = 1$, \dots, $m$ one has $\alpha e_{2,3} = 0$, so $\alpha = 0$. Then, \[ f = \sum_{l=2}^{m} \beta_l [y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}]z_{j_1}=0.\]
Suppose that there exists $l_0$ such that $\beta_{l_0} \neq 0$, and consider the evaluation $y_{h_{l_0}} = e_{1,2}$, $y_{h_i} = e_{2,2}$ for $i \neq l_0$ and $z_{j_1} = e_{2,3}$. Then, $\beta_{l_0}e_{1,3} = 0$, therefore $\beta_{l_0} = 0$, a contradiction. Thus, we have the desired result.
\end{proof}
\section{ Graded polynomial identities of $\mathcal{A}^3$}
By direct computation, the following polynomials are identities of $\mathcal{A}^3$:
\begin{itemize}
\item[(a)] $ z_{j_1}z_{j_2} $,
\item[(b)] $ [y_{i_1},y_{i_2}][y_{i_3},y_{i_4}]$,
\item[(c)] $ [y_{i_1},y_{i_2}]z_{j_1}$,
\item[(d)] $ [z_{j_1}[y_{i_1},y_{i_2}],y_{i_3}]$.
\end{itemize}
\begin{proposition} \label{pr:A_3}
The polynomial \begin{equation}
[z_{j_1},y_{i_3}][y_{i_1},y_{i_2}] + z_{j_1}[y_{i_1},y_{i_2},y_{i_3}]
\end{equation}
is a polynomial identity of the algebra $\mathcal{A}^3$.
\end{proposition}
\begin{proof}
Note that \begin{align*}
[z_{j_1},y_{i_3}] [y_{i_1},y_{i_2}] & = z_{j_1}y_{i_3}[y_{i_1},y_{i_2}] - y_{i_3}z_{j_1} [y_{i_1},y_{i_2}]
\stackrel{(d)}{=}z_{j_1}y_{i_3}[y_{i_1},y_{i_2}] - z_{j_1} [y_{i_1},y_{i_2}]y_{i_3} \\
& = z_{j_1}[y_{i_3},[y_{i_1},y_{i_2}]] = -z_{j_1}[y_{i_1},y_{i_2},y_{i_3}].\qedhere
\end{align*}
\end{proof}
Consider the vector space $P_{m,n}$ of multilinear polynomials in the variables $y_{i_1}$, $y_{i_2}$, \dots, $y_{i_m}$, $z_{j_1}, z_{j_2}, \dots, z_{j_n}$. Since $ z_{j_1}z_{j_2} = 0$, $P_{m,n} = 0 $ if $n\geq 2$. So, we consider the cases $n=0$ and $n=1$.
\subsection{Case $n=0$}
Since $[y_{i_1},y_{i_2}][y_{i_3},y_{i_4}] =0$,
then considering proper polynomials, we have only sums of commutators, and by Jacobi identity we can reorder the variables of that commutator and consider only the case $[ y_{i_1}, y_{i_2},\dots, y_{i_{m}} ]$ where $i_1 > i_2 < i_3 < \cdots < i_{m-2}$.
\begin{proposition}
The polynomials $[y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}]$ with $h_{l} > h_1 < h_2 < \cdots < h_m$ are linearly independent modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^3)$.
\end{proposition}
\begin{proof}
Consider the polynomial given by \[ f = \sum_{l=2}^{m}\beta_l [y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}]\] and suppose that $f\in \Id_{\mathbb{Z}_2}(\mathcal{A}^3)$.
Suppose there exists $l_0$ such that $\beta_{l_0} \neq 0$, and consider the evaluation $y_{h_{l_0}} = e_{2,3}$ and $y_{h_i} = (e_{1,1} + e_{3,3})$ for $i \neq l_0$. Then, $\beta_{l_0}e_{2,3} = 0$, therefore $\beta_{l_0} = 0$, a contradiction. Thus, we have the desired result.
\end{proof}
\subsection{Case $n=1$}
We work with proper polynomials.
Since $ [y_{i_1},y_{i_2}] [y_{i_3},y_{i_4}] = 0$ and we have only one $z_j$, then we only consider sums of commutators with products of two commutators.
As $ [y_{i_1},y_{i_2}]z_{j_1}=0$, when we have products of two commutators then $z_{j_1}$ is in the first commutator.
From the identity (c) $ [y_{i_1},y_{i_2},z_{j_1}] = - z_{j_1}[y_{i_1},y_{i_2}]$ and using the identity (d) we have $ [y_{i_1},y_{i_2},z_{j_1},y_{i_3}] = 0$. Therefore, if $z_{j_1}$ lies in a commutator $\omega$, that commutator has the form $\omega = [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_k}] $ or $\omega = [y_{i_1},y_{i_2}, \dots,y_{i_k},z_{j_1}] = -z_{j_1} [y_{i_1},y_{i_2}, \dots,y_{i_k}]$.
Let us consider the case $\omega = [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_k}] $. Note that by Jacobi identity $ [z_{j_1},y_{i_1},y_{i_2}] + [y_{i_1},y_{i_2},z_{j_1}] + [y_{i_2},z_{j_1},y_{i_1}] = 0 $, so $ [z_{j_1},y_{i_1},y_{i_2}] - z_{j_1}[y_{i_1},y_{i_2}] - [z_{j_1},y_{i_2},y_{i_1}] = 0 $ and
$ [z_{j_1},y_{i_2},y_{i_1}] = [z_{j_1},y_{i_1},y_{i_2}] - z_{j_1}[y_{i_1},y_{i_2}]$. Therefore, we can write the variables $y_i$ in any order in $\omega$.
If $\omega = [y_{i_1},y_{i_2}, \dots,y_{i_k},z_{j_1}] = -z_{j_1} [y_{i_1},y_{i_2}, \dots,y_{i_k}]$, as in the case $P_{m,0}$, we can reorder the variables $y_{i_1},y_{i_2}, \dots,y_{i_k}$ such that $z_{j_1}[y_{i_1},y_{i_2}, \dots,y_{i_k}]$ is a linear combination of polynomials $z_{j_1}[y_{i_l},y_{i_1},y_{i_2}, \dots ,\widehat{y_{i_l}}, \dots,y_{i_s}]$ where $i_{l} > i_1 < i_2 < \cdots < i_k$.
A nonzero product of commutators has the form $ [z_{j_1},y_{i_1},\dots,y_{i_s}] [y_{h_1},y_{h_2},\dots,y_{h_t}]$ and by Proposition \ref{pr:A_3} we reduce it to the form $ z_{j_1}[y_{i_1},y_{i_2},\dots,y_{i_{s+t}}]$.
Then, if $\mathcal{B}$ is the space of proper polynomials in the variables $z_{j_1}$, $y_{i_1}$, $y_{i_2}$, \dots, $y_{i_k}$, modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^3)$, $\mathcal{B}$ is spanned by \begin{itemize}
\item $ [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_k}]$, $i_1 < i_2 < \cdots < i_k$,
\item $ z_{j_1}[y_{i_l},y_{i_1},y_{i_2}, \dots ,\widehat{y_{i_l}}, \dots,y_{i_k}]$, $i_{l} > i_1 < i_2 < \cdots < i_k$.
\end{itemize}
\begin{theorem}\label{th:A_3}
The following identities form a basis for the $\mathbb{Z}_2$-graded identities for the algebra $\mathcal{A}^3$:
\begin{itemize}
\item[$(i)$] $ z_{i_1}z_{i_2}$,
\item[$(ii)$] $ [y_{i_1},y_{i_2}][y_{i_3},y_{i_4}] $,
\item[$(iii)$] $ [y_{i_1},y_{i_2}]z_{i_1} $,
\item[$(iv)$] $ [z_{j_1}[y_{i_1},y_{i_2}],y_{i_3}]$.
\end{itemize}
\end{theorem}
\begin{proof}
Let us show that the polynomials $ [z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_m}]$ with $i_1 < i_2 < \cdots < i_m$, and $ z_{j_1}[y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}] $ with $h_{l} > h_1 < h_2 < \cdots < h_m$ are linearly independent modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^3)$. Consider $f$ the polynomial given by \[ f = \alpha[z_{j_1},y_{i_1},y_{i_2}, \dots,y_{i_m}] + \sum_{l=2}^{m}\beta_l z_{j_1}[y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}] \] and suppose that $f\in \Id_{\mathbb{Z}_2}(\mathcal{A}^3).$ Now, considering the evaluation $z_{j_1} = e_{1,2}$ and $y_{i} = (e_{1,1} + e_{3,3})$ for $i = 1,\dots,m$ one has $\alpha e_{1,2} = 0$, so $\alpha = 0$. Then, \[ f = \sum_{l=2}^{m} \beta_l z_{j_1}[y_{h_l},y_{h_1},y_{h_2}, \dots ,\widehat{y_{h_l}}, \dots,y_{h_m}]=0.\]
Suppose there exists $l_0$ such that $\beta_{l_0} \neq 0$, and consider the evaluation $y_{h_{l_0}} = e_{2,3}$, $y_{h_i} = (e_{1,1}+e_{3,3})$ for $i \neq l_0$ and $z_{j_1} = e_{1,2}$. Then, $\beta_{l_0}e_{1,3} = 0$, therefore $\beta_{l_0} = 0$, a contradiction. Thus, we have the desired result.
\end{proof}
\section{$S_m \times S_n$-characters of $\mathcal{A}^1$}
In this section, we consider the cocharacters of the $\mathbb{Z}_2$-graded algebra $\mathcal{A}^1$ (i.e., the algebra $A$ with the grading given by (\ref{eq:gra1}). We recall that a basis for the graded identities of $\mathcal{A}^1$ is given by the polynomials $[y_1,y_2]$ and $z_1z_2z_3$.
Let $S_m$ act on the variables $y_1$, \dots, $y_m$ and let $S_n$ act on $z_1$, \dots, $z_n$, and consider the cocharacter
\[ \chi_{m,n}(\mathcal{A}^1) = \sum_{(\lambda, \mu) \vdash (m,n) } m_{\lambda,\mu}\chi_{\lambda} \otimes \chi_{\mu}.
\]
Let $\lambda \vdash m$, $\mu \vdash n$, and let $W_{\lambda,\mu}$ be a left irreducible $S_m \times S_n$-module. If $T_{\lambda}$ is a tableau of shape $\lambda$ and $T_{\mu}$ a tableau of shape $\mu$, then $W_{\lambda,\mu} \cong F(S_m \times S_n)e_{T_\lambda}e_{T_\mu}$ with $S_m$ and $S_n$ acting on disjoint sets of integers.
From the identity $[y_1, y_2] = 0$ we have $m_{(m),\emptyset} = 1$ and $m_{\lambda,\emptyset} = 0$ for $\lambda \neq (m)$. Also, from $[y_1, y_2] = 0$, we obtain $m_{\lambda,\mu} = 0$ for $h(\lambda) >2$.
Consider the case $n=1$. Since $P_{m,1}(\mathcal{A}^1) = \Span\{ y_{i_1}y_{i_2} \cdots y_{i_s} z y_{j_1}y_{j_2} \cdots y_{i_t} \}$, we consider the case $\lambda \vdash m$ with $\lambda = (p+q,p)$.
For every $i=0$, \dots, $q$ define the following two tableaux: $T_\lambda^{(i)}$ is the tableau
$$\begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|}
\hline$i+1$ & $i+2$ & $\cdots$ & $i+p$ & 1 & 2 & $\cdots$ & $i$ & $i+2 p+2$ & $\cdots$ & $2p+q+1$ \\
\hline$i+p+2$ & $i+p+3$ & $\cdots$ & $i+2 p+1$ & \multicolumn{7}{|c}{} \\
\cline { 1 - 4 }
\end{tabular},
$$
$$
T_\mu^{(i)}= \begin{tabular}{|l|}
\hline $i+p+1$ \\
\hline
\end{tabular}.
$$
The associated polynomial is
\begin{equation}\label{eq:yzy}
a_{p,q}^{(i)}(y_1,y_2,z) = y_1^i \underbrace{\bar{y}_1 \cdots \tilde{y}_1}_p z \underbrace{\bar{y}_2 \cdots \tilde{y}_2}_p y_1^{q-i},
\end{equation}
where $\bar{}$ and $\tilde{}$ mean alternation on the corresponding elements. That is,
\[ f(\overline{x}_1,\dots,\overline{x}_i, x_{i+1}, \dots, x_n) = \sum_{\sigma\in S_i} (-1)^{\sigma} f(x_{\sigma(1)},\dots, x_{\sigma(i)}, x_{i+1} , \dots , x_n).\]
\begin{proposition} \label{pro:linear_inde}
Modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$, the polynomials $a_{p,q}^{(i)}(y_1,y_2,z)$ as in (\ref{eq:yzy}) are linearly independent, $i = 0$, \dots, $q$.
\end{proposition}
\begin{proof}The proof is similar to that of Theorem 3 in \cite{valenti2002graded}.
Suppose that the polynomials $a_{p,q}^{(i)}$ are linearly dependent. Modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$, there exist $\alpha_i$, $i=0$, \dots, $q$ such that
\[ \sum_{i=0}^{q}\alpha_i a_{p,q}^{(i)}(y_1,y_2,z) = 0. \]
Let $t = \max\{ i \mid \alpha_i \neq 0 \}$. Then
\begin{equation} \label{eq:alpha_t}
\alpha_t a_{p,q}^{(t)}(y_1,y_2,z) + \sum_{i < t}\alpha_i a_{p,q}^{(i)}(y_1,y_2,z) = 0.
\end{equation}
Considering $y_1 = y_1 + y_3$ in (\ref{eq:alpha_t}), we have that
$ \begin{aligned}
& \alpha_t\left(y_1+y_3\right)^t \overline{\left(y_1+y_3\right)} \cdots\widetilde{\left(y_1+y_3\right)} z \bar{y}_2 \cdots \tilde{y}_2\left(y_1+y_3\right)^{q-t} \\
& \quad +\sum_{i<t} \alpha_i\left(y_1+y_3\right)^i \overline{\left(y_1+y_3\right)} \cdots\widetilde{\left(y_1+y_3\right)} z \bar{y}_2 \cdots \tilde{y}_2\left(y_1+y_3\right)^{q-i} = 0.
\end{aligned}$
Let us consider the homogeneous component of degree $t + p$ in the variable $y_1$ and degree $q - t$ in the variable $y_3$,
\[ \alpha_t y_1^{t}\bar{y}_1\cdots\tilde{y}_1 z \bar{y}_2\cdots\tilde{y}_2 y_3^{q-t} + \cdots = 0. \]
Substituting $y_1 = e_{1,1} + e_{3,3}$, $y_2 = y_3 = e_{2,2}$ and $z = e_{1,2}$, we obtain $\alpha_t\cdot e_{1,2} = 0$, so $\alpha_t = 0$ and we have the desired result.
\end{proof}
\begin{proposition} \label{pro:m_(1)}
$m_{\lambda,(1)} = q + 1$, if $\lambda = (p+q,p) \vdash m$.
\end{proposition}
\begin{proof}
For every $i$, $e_{T_{\lambda}^{(i)}}e_{T_{\mu}^{(i)}}(y_1,\dots,y_{m},z)$ is the complete linearization of the polynomial $a_{p,q}^{(i)}(y_1,y_2,z)$. By Proposition \ref{pro:linear_inde} $m_{\lambda,(1)} \geq q+1$ if $\lambda = (p+q,p) \vdash m$. Also, given $T_{\lambda}$ and $T_{\mu}$ two tableaux and $f = e_{T_{\lambda}^{(i)}}e_{T_{\mu}^{(i)}}(y_1,\dots,y_{m},z) \notin \Id_{\mathbb{Z}_2}(\mathcal{A}^1)$, any two alternating variables in $f$ must lie on different sides of $z$. Since $f$ is a linear combination (modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$) of polynomials, each alternating on $p$ pairs of $y_i$'s, we have that $f$ is a linear combination (modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$) of the polynomials $e_{T_{\lambda}^{(i)}}e_{T_{\mu}^{(i)}}$. Hence $m_{\lambda,(1)} = q + 1$, if $\lambda = (p+q,p) \vdash m$.
\end{proof}
Now, consider the case $n=2$, $P_{m,2}(\mathcal{A}^1) = \Span\{ y_{i_1}y_{i_2} \cdots y_{i_s} z_{l_1} y_{j_1}y_{j_2} \cdots y_{i_t} z_{l_2} \}$.
For $\mu \vdash 2$ we have the possibilities $\mu = (2)$ and $\mu = (1,1)$. Define the tableaux $T_{\lambda}^{(i)}$ as
$$\begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|}
\hline$i+1$ & $i+2$ & $\cdots$ & $i+p$ & 1 & 2 & $\cdots$ & $i$ & $i+2 p+2$ & $\cdots$ & $2p+q+1$ \\
\hline$i+p+2$ & $i+p+3$ & $\cdots$ & $i+2 p+1$ & \multicolumn{7}{|c}{} \\
\cline { 1 - 4 }
\end{tabular}
$$
and
$$
T_{(2)}^{(i)}= \begin{tabular}{|l|l|}
\hline $i+p+1$ & $2p +q+ 2$ \\
\hline
\end{tabular}, \quad
T_{(1,1)}^{(i)}= \begin{tabular}{|l|}
\hline $i+p+1$ \\
\hline $2p +q+ 2$ \\
\hline
\end{tabular}
$$
Then, the associated polynomials are
\begin{equation} \label{eq:yzyz}
a_{p,q}^{(i)}(y_1,y_2,z) = y_1^i \underbrace{\bar{y}_1 \cdots \tilde{y}_1}_p z \underbrace{\bar{y}_2 \cdots \tilde{y}_2}_p y_1^{q-i}z,
\end{equation}
and respectively
\begin{equation}\label{eq:yzyz2}
a_{p,q}^{(i)}(y_1,y_2,z_1,z_2) = y_1^i \underbrace{\bar{y}_1 \cdots \tilde{y}_1}_p \check{z}_1 \underbrace{\bar{y}_2 \cdots \tilde{y}_2}_p y_1^{q-i} \check{z}_2,
\end{equation}
First we consider the case of the polynomials (\ref{eq:yzyz}).
\begin{proposition} \label{pro:linear_inde2}
Modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$, the polynomials $a_{p,q}^{(i)}(y_1,y_2,z)$ as in (\ref{eq:yzyz}) are linearly independent, $i = 0$, \dots, $q$.
\end{proposition}
\begin{proof}
Suppose that the polynomials $a_{p,q}^{(i)}$ are linearly dependent. Then, modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$ there exist $\alpha_i$, $i=0$, \dots, $q$ such that
\[ \sum_{i=0}^{q}\alpha_i a_{p,q}^{(i)}(y_1,y_2,z) = 0. \]
Let $t = \max\{ i \mid \alpha_i \neq 0 \}$. Then,
\begin{equation} \label{eq:alpha_t2}
\alpha_t a_{p,q}^{(t)}(y_1,y_2,z) + \sum_{i < t}\alpha_i a_{p,q}^{(i)}(y_1,y_2,z) = 0.
\end{equation}
Considering $y_1 = y_1 + y_3$ in (\ref{eq:alpha_t2}), we have that
$ \begin{aligned}
& \alpha_t\left(y_1+y_3\right)^t \overline{\left(y_1+y_3\right)} \cdots\widetilde{\left(y_1+y_3\right)} z \bar{y}_2 \cdots \tilde{y}_2\left(y_1+y_3\right)^{q-t}z \\
& \quad +\sum_{i<t} \alpha_i\left(y_1+y_3\right)^i \overline{\left(y_1+y_3\right)} \cdots\widetilde{\left(y_1+y_3\right)} z \bar{y}_2 \cdots \tilde{y}_2\left(y_1+y_3\right)^{q-i}z = 0.
\end{aligned}$
We consider the homogeneous component of degree $t + p$ in $y_1$ and of degree $q - t$ in $y_3$,
\[ \alpha_t y_1^{t}\bar{y}_1\cdots\tilde{y}_1 z \bar{y}_2\cdots\tilde{y}_2 y_3^{q-t}z + \cdots = 0. \]
Substituting $y_1 = e_{1,1} + e_{3,3}$, $y_2 = y_3 = e_{2,2}$ and $z = e_{1,2} + e_{2,3}$, we obtain $\alpha_t\cdot e_{1,3} = 0$, so $\alpha_t = 0$ and we have the result.
\end{proof}
From Proposition \ref{pro:linear_inde2} and with a similar argument to Proposition \ref{pro:m_(1)}, we obtain
\begin{proposition} \label{pro:m_(2)}
$m_{\lambda,(2)} = q + 1$, if $\lambda = (p+q,p) \vdash m$.
\end{proposition}
Finally, we consider the case of the polynomials (\ref{eq:yzyz2}),
\[ a_{p,q}^{(i)}(y_1,y_2,z_1,z_2) = y_1^i \underbrace{\bar{y}_1 \cdots \tilde{y}_1}_p \check{z}_1 \underbrace{\bar{y}_2 \cdots \tilde{y}_2}_p y_1^{q-i} \check{z}_2. \]
\begin{proposition} \label{pro:linear_inde3}
Modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$, the polynomials $a_{p,q}^{(i)}(y_1,y_2,z_1,z_2)$ as in (\ref{eq:yzyz2}) are linearly independent, $i = 0$, \dots, $q$.
\end{proposition}
\begin{proof}
If we consider the evaluation $z_1 = e_{1,2}$ and $z_2 = e_{2,3}$, then
\[ a_{p,q}^{(i)}(y_1,y_2,z_1,z_2) = y_1^i \underbrace{\bar{y}_1 \cdots \tilde{y}_1}_p z_1 \underbrace{\bar{y}_2 \cdots \tilde{y}_2}_p y_1^{q-i} z_2. \]
Suppose the polynomials $a_{p,q}^{(i)}$ are linearly dependent. Modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$ there exist $\alpha_i$, $i=0$, \dots, $q$ such that
\[ \sum_{i=0}^{q}\alpha_i a_{p,q}^{(i)}(y_1,y_2,z_1,z_2) = 0. \]
Let $t = \max\{ i \mid \alpha_i \neq 0 \}$. Then,
\begin{equation} \label{eq:alpha_t3}
\alpha_t a_{p,q}^{(t)}(y_1,y_2,z_1,z_2) + \sum_{i < t}\alpha_i a_{p,q}^{(i)}(y_1,y_2,z_1,z_2) = 0.
\end{equation}
Considering $y_1 = y_1 + y_3$ in (\ref{eq:alpha_t3}), we have that
$ \begin{aligned}
& \alpha_t\left(y_1+y_3\right)^t \overline{\left(y_1+y_3\right)} \cdots\widetilde{\left(y_1+y_3\right)} \check{z}_1 \bar{y}_2 \cdots \tilde{y}_2\left(y_1+y_3\right)^{q-t}\check{z}_2 \\
& \quad +\sum_{i<t} \alpha_i\left(y_1+y_3\right)^i \overline{\left(y_1+y_3\right)} \cdots\widetilde{\left(y_1+y_3\right)} \check{z}_1 \bar{y}_2 \cdots \tilde{y}_2\left(y_1+y_3\right)^{q-i}\check{z}_2 = 0.
\end{aligned}$
The homogeneous component of degree $t + p$ in $y_1$ and of degree $q - t$ in $y_3$ is
\[ \alpha_t y_1^{t}\bar{y}_1\cdots\tilde{y}_1 \check{z}_1 \bar{y}_2\cdots\tilde{y}_2 y_3^{q-t} \check{z}_2 + \cdots = 0. \]
Substituting $y_1 = e_{1,1} + e_{3,3}$, $y_2 = y_3 = e_{2,2}$, $z_1 = e_{1,2}$ and $ z_2 = e_{2,3}$, we obtain $\alpha_t\cdot e_{1,3} = 0$, so $\alpha_t = 0$ and we are done.
\end{proof}
\begin{proposition} \label{pro:m_(1,1)}
$m_{\lambda,\mu} = q + 1$, if $\lambda = (p+q,p) \vdash m$ and $\mu = (1,1)$.
\end{proposition}
\begin{proof}
For every $i$, $e_{T_{\lambda}^{(i)}}e_{T_{\mu}^{(i)}}(y_1,\dots,y_{m},z_1,z_2)$ is the complete linearisation of the element $a_{p,q}^{(i)}(y_1,y_2,z_1,z_2)$. By Proposition \ref{pro:linear_inde3}, $m_{\lambda,\mu} \geq q+1$ if $\lambda = (p+q,p) \vdash m$ and $\mu = (1,1)$. Also, given $T_{\lambda}$ and $T_{\mu}$, and $f = e_{T_{\lambda}^{(i)}}e_{T_{\mu}^{(i)}}(y_1,\dots,y_{m},z_1,z_2) \notin \Id_{\mathbb{Z}_2}(\mathcal{A}^1)$, any two alternating variables in $f$ must lie on different sides of $z_1$ or $z_2$, or the two alternating variables are $z_1$ and $z_2$. Since $f$ is a linear combination (modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$) of polynomials, each alternating on $p$ pairs of $y_i$'s, and alternating on $z_i$'s, we have that $f$ is a linear combination (modulo $\Id_{\mathbb{Z}_2}(\mathcal{A}^1)$) of the polynomials $e_{T_{\lambda}^{(i)}}e_{T_{\mu}^{(i)}}$. Hence $m_{\lambda,(1,1)} = q + 1$, if $\lambda = (p+q,p) \vdash m$.
\end{proof}
Finally, based on the previous results, we have the following theorem on the graded cocharacters of the algebra $\mathcal{A}^1$.
\begin{theorem}
Let
\[ \chi_{m,n}(\mathcal{A}^1) = \sum_{(\lambda, \mu) \vdash (m,n) } m_{\lambda,\mu}\chi_{\lambda} \otimes \chi_{\mu} \]
be the $(m,n)$-cocharacter of $\mathcal{A}^1$. Then
\begin{itemize}
\item[(1)] $m_{\lambda,\emptyset} = 1$, if $\lambda = (m) \vdash m$;
\item[(2)] $m_{\lambda,(1)} = q+1$, if $\lambda = (p+q,p) \vdash m$;
\item[(3)] $m_{\lambda,\mu} = q+1$, if $\lambda = (p+q,p) \vdash m$ and $\mu \vdash 2$.
\end{itemize}
In all remaining cases $m_{\lambda,\mu} = 0$.
\end{theorem}
\section{Identities with involutions on $A$} \label{InvotionOnA}
We are interested in studying the identities of the algebra $A$, considering it as an algebra with involution. We can define on the algebra $A$ the involution $*$ obtained by reflecting a matrix along its secondary diagonal, that is \[ \begin{pmatrix} d & a & c \\ 0 & g & b \\ 0 & 0 & d \end{pmatrix}^{*} = \begin{pmatrix} d & b & c \\ 0 & g & a \\ 0 & 0 & d \end{pmatrix}. \]
Then,
\[ A^{+} = \begin{pmatrix} d & a & c \\ 0 & g & a \\ 0 & 0 & d \end{pmatrix} \text{ and } A^{-} = \begin{pmatrix} 0 & a & 0 \\ 0 & 0 & -a \\ 0 & 0 & 0 \end{pmatrix}.
\]
This involution was studied in \cite{MISHCHENKO200066}.
Denote the free algebra with involution by $K\langle X, * \rangle = K\langle Y \cup Z \rangle$ generated by symmetric and skew elements, that is \[ K\langle Y \cup Z \rangle = K \langle y_{1}, z_{1}, y_{2}, z_{2}, , \dots \rangle \]
where $y_{i}$ stands for a symmetric variable and $z_{i}$ for a skew-symmetric variable.
If in a polynomial it is allowed some variable to be either $y_i$ or $ z_i$, we denote it as $x_i$, and we set $|x_i|=0$ if $x_i=z_i$ and $|x_i|=1$ when $x_i=y_i$. Put $|x_i x_j| = 0$ when the commutator $[x_i, x_j]$ is skew-symmetric, and $|x_i x_j|= 1$ if the commutator $[x_i, x_j]$ is symmetric.
The following theorem describes a basis of $\Id^{*}(A)$.
\begin{theorem}
The $T^{*}$-ideal $\Id^{*}(A)$ is generated by the following polynomials:
\begin{tabular}{llll} \label{th:Base_*}
1. $z_1y_1z_2y_2z_3$ & 2. $[z_1,z_2]$ & 3. $[z_{1}y_1z_{2},y_2]$ & 4. $z_1y_1z_2 - z_2y_1z_1$ \\
\multicolumn{4}{l}{ 5. $(-1)^{|x_1x_2|}[x_1,x_2][x_3,x_4] - (-1)^{|x_3x_4|}[x_3,x_4][x_1,x_2]$} \\
\multicolumn{4}{l}{ 6. $(-1)^{|x_1x_2|}[x_1,x_2][x_3,x_4] - (-1)^{|x_1x_3|}[x_1,x_3][x_2,x_4] + (-1)^{|x_1x_4|}[x_1,x_4][x_2,x_3]$} \\
\multicolumn{4}{l}{ 7. $ [x_1,x_2]z_5[x_3,x_4]$}
\end{tabular}
\end{theorem}
The proof of Theorem \ref{th:Base_*} will be a consequence of the following propositions.
Let $\mathcal{I}$ be the $T^{*}$-ideal generated by the identities (1) to (7) of Theorem \ref{th:Base_*}.
\begin{proposition} \label{pr:order*_m=2}
The following is an identity modulo $\mathcal{I}$
\begin{equation}
y_{\sigma(i_1)}y_{\sigma(i_2)}\cdots y_{\sigma(i_s)}z_1y_{\rho(j_1)}y_{\rho(j_2)}\cdots y_{\rho(j_t)}z_2 = y_{i_1}y_{i_2}\cdots y_{i_s}z_1y_{j_1}y_{j_2}\cdots y_{j_t}z_2,
\end{equation}
for $\sigma \in S_s$ and $\rho \in S_t$.
\end{proposition}
\begin{proof}
Consider the case $y_{i_2}y_{i_1}z_1y_{j_2}y_{j_1}z_2$ and note that $[y_1,y_2] \in A^{-}$. Then,
\begin{align*}
y_{i_2}y_{i_1}z_1y_{j_2}y_{j_1}z_2 & = y_{i_1}y_{i_2}z_1y_{j_2}y_{j_1}z_2 + [y_{i_2},y_{i_1}]z_1y_{j_2}y_{j_1}z_2 = y_{i_1}y_{i_2}z_1y_{j_2}y_{j_1}z_2 \\
& = y_{i_1}y_{i_2}z_1y_{j_1}y_{j_2}z_2 + y_{i_1}y_{i_2}z_1[y_{j_2},y_{j_1}]z_2 = y_{i_1}y_{i_2}z_1y_{j_1}y_{j_2}z_2.
\end{align*}
It follows that \[
y_{\sigma(i_1)}y_{\sigma(i_2)}\cdots y_{\sigma(i_s)}z_1y_{\rho(j_1)}y_{\rho(j_2)}\cdots y_{\rho(j_t)}z_2 = y_{i_1}y_{i_2}\cdots y_{i_s}z_1y_{j_1}y_{j_2}\cdots y_{j_t}z_2,\]
for $\sigma \in S_s$ and $\rho \in S_t$.
\end{proof}
Let $P_{n,m}$ be the set of multilinear polynomials in $n$ symmetric variables and $m$ skew-symmetric variables. By Id. 1., $P_{n,m} = 0$ if $m>2$. Then, we consider the cases $m = 0$, $1$, $2$ separately.
\subsection{Case $m = 0$}
Let $\Gamma_{n,0}(\mathcal{I})$ be the subspace of the $Y$-proper multilinear polynomials in the variables $y_1$, \dots, $y_m$ of the relatively free algebra $F\langle Y,Z, * \rangle/\mathcal{I}$.
Since $[x_1, x_2][x_3, x_4][x_5, x_6]$ belongs to $\mathcal{I}$, the vector
space $\Gamma_{n,0}(\mathcal{I})$ is spanned by the proper polynomials \[ [x_{j_1}, \dots , x_{j_s}] [x_{k_1}, \dots , x_{k_t}], \]
where $s$, $t \geq 0$, $s \neq 1$, $t \neq 1$, $j_1 > j_2 < \cdots < j_s$, and $k_1 > k_2 < \cdots < k_t$.
Note that $A^{+} = UT_3(K)^{+}$. We use a result of \cite{di2006involutions}.
\begin{definition}
A polynomial $f$ is called {\sl $S_3$-standard} if $f$ is either of type $[y_{j_1}, \dots , y_{j_n}]$ or $[y_{j_1}, \dots , y_{j_{n-2}}][y_{k_1}, y_{k_2}]$, where the commutator $[y_{i_1}, \dots , y_{i_s}]$ satisfies $i_1 > i_2 < \cdots < i_s$, and if $f$ is of the second type we have that $j_1 >k_1$, ${j_2} >k_2$.
\end{definition}
The next result is a particular case of Proposition 5.8 and Lemma 6.2 of \cite{di2006involutions}.
\begin{proposition} \label{pro:base*_m=0}
The following statements hold:
\begin{itemize}
\item[$(i)$] The $S_3$-standard polynomials span the vector space $\Gamma_{n,0}(A)$.
\item[$(ii)$] The $S_3$-standard polynomials in $\Gamma_{n,0}$ are linearly independent modulo the ideal $\Id^{*}(A)$.
\end{itemize}
\end{proposition}
\subsection{Case $m = 1$}
First, we describe a spanning set of $ P_{n,1}(A)$.
\begin{proposition}\label{pr:spn*_m=1}
$ P_{n,1}(A)$ is spanned by elements of the form
\begin{itemize}
\item[$(i)$] $ y_{i_1}y_{i_2}\cdots y_{i_s}zy_{i_{s+1}}y_{i_{s+2}}\cdots y_{i_n}$, with $i_1<\cdots<i_s$ and $i_{s+1} < \cdots < i_n$;
\item[$(ii)$] $ y_{i_1}y_{i_2}\cdots y_{i_s}zy_{i_{s+1}}y_{i_{s+2}}\cdots y_{i_{n-2}}[y_{i_{n-1}},y_{i_{n}}]$, with $i_1<\cdots<i_s$, $i_{s+1} < \cdots < i_{n-1}$ and $i_{n}<i_{n-1}$.
\end{itemize}
\end{proposition}
\begin{proof}
Let $ \omega \in P_{n,1}(A)$ be a monomial. Then, $\omega = y_{i_1}y_{i_2}\cdots y_{i_s}zy_{j_1}y_{j_2}\cdots y_{j_t}$ with $s+t = n$.
Suppose $\omega = \omega_1 z \omega_2$, there are two cases for $\omega_2$:
Case 1. {\sl $\omega_2$ with indices in non-increasing order}.
Consider $\omega = \omega_1 z \omega_2$ with $\omega_2 = w_{2,1}y_{l_2}y_{l_1}w_{2,2}$ and $y_{l_2},y_{l_1}$ the first pair of variables such that $l_2 > l_1$.
We can rewrite $\omega_2$ as $ \omega_2 = w_{2,1}y_{l_1}y_{l_2}w_{2,2} + w_{2,1}[y_{l_2},y_{l_1}]w_{2,2}$, then
\[
\omega = \omega_1 z w_{2,1}y_{l_1}y_{l_2}w_{2,2} + \omega_1 z w_{2,1}[y_{l_2},y_{l_1}]w_{2,2}
= \omega_1 z \overline{\omega}_2 + \omega_1 w_{2,2} z w_{2,1}[y_{l_2},y_{l_1}].
\]
Note that by Proposition \ref{pr:order*_m=2}, in the element $\omega_1 w_{2,2} z w_{2,1}[y_{l_2},y_{l_1}] $ we can reorder the variables of $\omega_1 w_{2,2}$ and $ w_{2,1}$ (separately) with the condition that if $ w_{2,1} = y_{k_1}\dots y_{k_p}$, then $y_{k_1} < \cdots < y_{k_p} < y_{l_2}$ and $ y_{l_1} < y_{l_2}$.
We repeat the process for $\omega_1z\overline{\omega}_2$ until getting an ascending order in the indices of the $y$'s to the right side of $z$ in the element without commutator.
Thus, $\omega = \omega_1 z \omega_2$ can be written as \[ \omega = \omega_1 z y_{h_1}\cdots y_{h_r} + \sum_{I,J} \alpha_{I,J} y_{i_1}\cdots y_{i_s} z y_{j_1}\cdots y_{j_{t-2}}[y_{j_{t-1}},y_{j_t}] \] with ${h_1}<\cdots <{h_r}$, ${i_1}<\cdots < {i_s}$, ${j_1}<\cdots< y_{j_{t-1}}$, and $j_t < {j_{t-1}}$.
If the indices of the variables of $\omega_1$ are in increasing order, then we have $\omega$ in the desired form.
Case 2. {\sl $\omega_2$ with indices in increasing order}.
Let $\omega = \omega_1 z \omega_2$ with $\omega_1 = w_{1,1}y_{l_2}y_{l_1}w_{1,2}$, and let $y_{l_2},y_{l_1}$ be the first pair of variables such that $l_2 > l_1$ and the indices of the variables of $\omega_2$ increase.
We rewrite $\omega_1$ as $ \omega_1 = w_{1,1}y_{l_1}y_{l_2}w_{1,2} + w_{1,1}[y_{l_2},y_{l_1}]w_{1,2}$, then
\[
\omega = w_{1,1}y_{l_1}y_{l_2}w_{1,2} z \omega_2 + w_{1,1}[y_{l_2},y_{l_1}]w_{1,2} z \omega_2 = \overline{\omega}_1z\omega_2 + w_{1,1}\omega_2 z w_{1,2} [y_{l_2},y_{l_1}].
\]
We apply once again Proposition \ref{pr:order*_m=2} to the element $ w_{1,1}\omega_2 z w_{1,2} [y_{l_2},y_{l_1}] $. Thus we reorder the variables of $w_{1,1}\omega_2$ and $ w_{1,2}$ (separately). Then, \[ \omega = \overline{\omega}_1z\omega_2 + \varpi_1 z \varpi_2 [y_{l_2},y_{l_1}], \] with the indices of the variables of $\omega_2$, $ \varpi_1$ and $\varpi_2$ in increasing order, respectively.
Note that there is no relation between the indices of the variables of $\varpi_2$ and those of the variables of $[y_{l_2},y_{l_1}]$. Take
\[ \omega = \overline{\omega}_1z\omega_2 + \varpi_1 z \varpi_2 y_{l_2}y_{l_1} - \varpi_1 z \varpi_2 y_{l_1}y_{l_2}. \]
For $ \varpi_1 z \varpi_2 y_{l_2}y_{l_1} $ and $ \varpi_1 z \varpi_2 y_{l_1}y_{l_2}$ we have two options:
\begin{itemize}
\item[$\bullet$] The indices of the variables of $\varpi_2 y_{l_2}y_{l_1}$ or $\varpi_2 y_{l_2}y_{l_1}$ are not in increasing order. In this case, we proceed as in Case 1 for $\varpi_1z\varpi_2 y_{l_2}y_{l_1}$ or $\varpi_1 z\varpi_2 y_{l_2}y_{l_1}$.
\item[$\bullet$] The indices of the variables of $\varpi_2 y_{l_2}y_{l_1}$ are in increasing order. In this case, we have the desired order for $ \varpi_1 z \varpi_2 y_{l_1}y_{l_2}$.
\end{itemize}
Finally, if necessary, we repeat the process for $ \overline{\omega}_1 z \omega_2$.
This process of rearranging the variables using cases 1. and 2., has finitely many steps, so eventually it ends.
\end{proof}
Now, we show that the polynomials of Proposition \ref{pr:spn*_m=1} are linearly independent.
\begin{proposition}\label{pr:base*_m=1}
The polynomials of the form
\begin{itemize}
\item[$(I)$] $ y_{i_1}y_{i_2}\cdots y_{i_s}zy_{i_{s+1}}y_{i_{s+2}}\cdots y_{i_n}$, with $i_1<\cdots<i_s$ and $i_{s+1} < \cdots < i_n$;
\item[$(II)$] $ y_{i_1}y_{i_2}\cdots y_{i_s}zy_{i_{s+1}}y_{i_{s+2}}\cdots y_{i_{n-2}}[y_{i_{n-1}},y_{i_{n}}]$, with $i_1<\cdots<i_s$, $i_{s+1} < \cdots < i_{n-1}$ and $i_{n}<i_{n-1}$,
\end{itemize}
are linearly independent modulo $\Id^{*}(A)$.
\end{proposition}
\begin{proof}
Let $f$ be a linear combination of elements of the form $(I)$ and $(II)$, such that $f$ is a $*$-polynomial identity of $A$.
Suppose that not all elements of $f$ of the form $(I)$ have non-zero coefficients, and consider $m = y_{l_1}\cdots y_{l_s}zy_{l_{s+1}}\cdots y_{l_n}$.
Perform the evaluation \begin{equation} \label{eq:ev1_m=1}
y_{l_1} = \cdots = y_{l_s} = e_{1,1} + e_{3,3}, \quad z = e_{1,2} - e_{2,3}, \quad y_{l_{s+1}} = \cdots = y_{l_n} = e_{2,2}.
\end{equation}
Since $e_{1,1} + e_{3,3}$ and $e_{2,2}$ commute, the elements with commutator in $f$ vanish.
Now, suppose there exists $\overline{m} = y_{j_1}\cdots y_{j_t}zy_{j_{t+1}}\cdots y_{j_n} $ of the form $(I)$ such that $\overline{m}$ is non-zero under the evaluation (\ref{eq:ev1_m=1}). Then, in $\overline{m}$ the $e_{2,2}$'s can only be substituted on one side of $z$ and similarly for the $(e_{1,1} + e_{3,3})$'s. Then we have two possibilities: $\{ j_1, \dots, j_t \} = \{ l_1, \dots l_s\} $ and $\{ j_{t+1}, \dots, j_n \} = \{ l_{s+1}, \dots l_n\}$, or $\{ j_1, \dots, j_t \} = \{ l_{s+1}, \dots l_n\} $, and $\{ j_{t+1}, \dots, j_n \} = \{ l_1, \dots l_s\} $. In the first case $m = \overline{m}$ and $m$ has zero coefficient in $f$. In the second case the evaluations of $m$ and $\overline{m}$ are $e_{1,2}$ and $e_{2,3}$, respectively. Therefore, $m$ and $\overline{m}$ have coefficient zero.
Suppose not all elements with one commutator have zero coefficient, and consider \[ m = y_{i_1}y_{i_2}\cdots y_{i_k}zy_{i_{k+1}}y_{i_{k+2}}\cdots y_{i_{n-2}}[y_{i_{n-1}},y_{i_{n}}] \] with the following properties:
\begin{itemize}
\item[$(i)$] its coefficient is non-zero,
\item[$(ii)$] the number $n-k-2$ of variables between $z$ and the commutator is the largest among all elements with non-zero coefficient,
\item[$(iii)$] $i_{n-1}$ is the least of all elements such that the properties $(i)$ and $(ii)$ hold.
\end{itemize}
Consider the evaluation \begin{equation} \label{eq:ev2_m=1}
y_{i_1} = \cdots = y_{i_k} = E, z = e_{1,2} - e_{2,3}, y_{i_{k+1}} = \cdots = y_{i_{n-1}} = e_{2,2}, y_{i_n} = e_{1,2}+e_{2,3}.
\end{equation}
Let $\overline{m} = y_{j_1}\cdots y_{j_s}zy_{j_{s+1}}\cdots y_{j_{n-2}}[y_{j_{n-1}},y_{j_{n}}] $ be a polynomial of the form $(II)$, such that $\overline{m}$ does not vanish under the substitution (\ref{eq:ev2_m=1}). We look for the conditions $\overline{m}$ must satisfy.
Since $E$ and $e_{2,2}$ commute, the elements to be substituted in the commutator will be $e_{1,2} + e_{2,3} $ and $e_{2,2}$. Also $e_{2,2}$ can only be substituted on the right side of $z = (e_{1,2}-e_{2,3})$, and since $n-k-2 \geq n-s-2$, then the $E$'s only can be substituted on the left side of $z$. Then $s = k$ and $\{ j_1, \dots, j_s \} = \{ i_1, \dots i_k\} $.
If $j_{n-1} \notin \{i_{n-1},i_{n}\}$ then $j_{n-1} \in \{ i_{k+1}, \dots ,i_{n-2} \}$, then $j_{n-1}<i_{n-1}$, but this contradicts the hypothesis of minimality of $i_{n-1}$. So $j_{n-1} \in \{i_{n-1},i_{n}\}$, and since $i_n < i_{n-1}$ we have $j_{n-1} = i_{n-1}$ and $j_n = i_n$. We conclude $\{ i_{k+1}, \dots, i_{n_2} \} = \{ j_{k+1}, \dots, j_{n_2} \} $, $m = \overline{m}$ and the coefficient of $m$ is zero.
\end{proof}
\subsection{Case $m = 2$}
By Proposition \ref{pr:order*_m=2} and the identities $3.$ and $4.$, $P_{n,2}(A)$ is spanned by monomials of the form \[ y_{i_1}y_{i_2}\cdots y_{i_s}z_1y_{j_1}y_{j_2}\cdots y_{j_t}z_2 \] with $i_1<i_2\cdots<i_s$, $j_1<j_2\cdots<j_t$ and $s+t = n$.
\begin{proposition} \label{pro:base*_m=2}
The polynomials
\begin{equation} \label{eq:*_m=2}
y_{i_1}y_{i_2}\cdots y_{i_s}z_1y_{j_1}y_{j_2}\cdots y_{j_t}z_2 \end{equation}
with $i_1<i_2\cdots<i_s$, $j_1<j_2\cdots<j_t$ and $s+t = n$, are linearly independent modulo $\Id(A)$.
\end{proposition}
\begin{proof}
Suppose that \[ f = \sum_{I,J}\alpha_{I,J} y_{i_1}y_{i_2}\cdots y_{i_s}z_1y_{j_1}y_{j_2}\cdots y_{j_t}z_2\] is a $*$-identity of $A$. Choose $\alpha_{I',J'} \neq 0$, such that $|J'|=t'$ is the largest possible. Now, consider the substitution $y_{i_1} = \cdots = y_{i_{s'}} = E$, $y_{j_1} = \cdots = y_{j_{t'}} = e_{2,2}$ and $z_1 = z_2 = (e_{1,2} - e_{2,3})$. Considering the order of the indices and that $|J'|$ is the largest, the only non-zero term of $f$ after the substitution will be the one with coefficient $\alpha_{I',J'}$. Hence $\alpha_{I',J'} = 0$, a contradiction. Thus, the polynomials in Equation (\ref{eq:*_m=2}) are linearly independent.
\end{proof}
Finally, by Proposition \ref{pro:base*_m=0}, Proposition \ref{pr:spn*_m=1}, Proposition \ref{pr:base*_m=1} and Proposition \ref{pro:base*_m=2} we have that Theorem \ref{th:Base_*} holds.
\section{Graded involution}
In this section we consider $A$ as a graded algebra with the grading as in (\ref{eq:gra1}), and the involution from Section \ref{InvotionOnA}. Then $*$ is a graded involution and \[ A_{0}^{+} = A_0, \qquad A_{0}^{-} = \{0\}, \qquad A_{1}^{+} = K(e_{1,2} + e_{2,3}), \qquad A_{1}^{-} = K(e_{1,2} - e_{2,3}).\]
\subsection{Graded $*$-polynomial identities of $A$}
Consider the free algebra with involution $K\langle Y \cup Z, * \rangle$, generated by symmetric and skew elements of even and odd degree, that is \[ K\langle Y \cup Z, * \rangle = K \langle y_{1}^{+}, y_{1}^{-}, z_{1}^{+}, z_{1}^{-}, y_{2}^{+}, y_{2}^{-}, z_{2}^{+}, z_{2}^{-}, \dots \rangle \]
where $y_{i}^{+}$ stands for a symmetric variable of even degree, $y_{i}^{-}$ for a skew variable of even degree, $z_{i}^{+}$ for a
symmetric variable of odd degree and $z_{i}^{-}$ for a skew variable of odd degree.
The following theorem gives us a basis of $\Id_{\mathbb{Z}_2}^{*}(A)$.
\begin{theorem} \label{th:Base_gr_*}
The $T_2^{*}$-ideal $\Id_{\mathbb{Z}_2}^{*}(A)$ is generated, as a $T_2^{*}$-ideal, by the following polynomials:
\begin{tabular}{lll}
1. $y^{-}$ & 2. $[y_{1}^{+},y_{2}^{+}]$ & 3. $[z_{1}^{+},z_{2}^{+}]$ \\
4. $[z_{1}^{-},z_{2}^{-}]$ & 5. $z_{1}^{+} \circ z_{1}^{-}$ & 6. $[y^{+},z_{1}z_{2}]$ \\
7. $[y_{1}^{+},z_{1}y_{2}^{+}z_{2}]$ & 8. $z_1 z_2 z_3$ & 9. $ z_{1}^{+}y^{+}z_{2}^{+} - z_{2}^{+}y^{+}z_{1}^{+}$ \\
10. $ z_{1}^{-}y^{+}z_{2}^{-} - z_{2}^{-}y^{+}z_{1}^{-}$ & 11. $ z_{1}^{-}y^{+}z_{2}^{+} + z_{2}^{+}y^{+}z_{1}^{-}$ &
\end{tabular}
\end{theorem}
\begin{proof}
One easily sees that the polynomials 1. -- 11. are graded $*$-identities. Let $P_{n_1,n_2,n_3,n_4}(A)$ be the set of multilinear graded $*$-polynomials modulo $\Id_{\mathbb{Z}_2}^{*}(A)$ in $n_1$ symmetric variables of degree $0$, $n_2$ skew variables of degree $0$, $n_3$ symmetric variables of degree $1$ and $n_1$ skew variables of degree $1$. From identities 1. and 8., we have that $P_{n_1,n_2,n_3,n_4}(A) = \{ 0 \}$ if $n_2>0$ or $n_3 + n_4 >2$.
\begin{enumerate}
\item[(i)] {\sl Case $n_3 + n_4 = 0$.} From identity 2., it is easy to see that \[ P_{n_1,0,0,0}(A) = \Span\{ y_1^{+}y_2^{+}\cdots y_{n_1}^{+} \}. \]
\item[(ii)] {\sl Case $n_3 + n_4 = 1$.} We consider the case $n_3 = 1$ (the case $n_4 = 1$ is analogous). Again, by identity 2., we see that \begin{equation} \label{eq:*_gr_z-1}
P_{n_1,0,1,0}(A) = \Span\{ y_{i_1}^{+} \cdots y_{i_s}^{+}z^{+}y_{j_1}^{+}\cdots y_{j_t}^{+} \},
\end{equation} with $s+t = n_1$, $i_1<\dots<i_s$ and $j_1 < \cdots< j_t$.
But the monomials $y_{i_1}^{+} \cdots y_{i_s}^{+}z^{+}y_{j_1}^{+}\cdots y_{j_t}^{+}$ are linearly independent. Indeed, let \[ f = \sum_{I,J}\alpha_{I,J}y_{i_1}^{+} \cdots y_{i_s}^{+}z^{+}y_{j_1}^{+}\cdots y_{j_t}^{+}.\]
Suppose $f\equiv 0$ modulo $Id_{\mathbb{Z}_2}^{*}(A)$ and that there are sets of indices $I_0$ and $J_0$, such that $\alpha_{I_0,J_0} \neq 0$. Considering the evaluation $y_i^{+} = e_{1,1} + e_{3,3}$, $y_{j}^{+} = e_{2,2}$, $z^{+} = e_{1,2} + e_{2,3}$, for $i \in I_0$, $j \in J_0$, we obtain $\alpha_{I_0,J_0}e_{1,2} = 0$. So, $\alpha_{I_0,J_0} = 0$, a contradiction. Then, the polynomials in (\ref{eq:*_gr_z-1}) are linearly independent.
\item[(iii)] {\sl Case $n_3 + n_4 = 2$.} Consider the case $n_3 = 2$ (the cases $n_4 = 2 $ and $n_3 = n_4 = 1$ are analogous). From the identities 2., 3., 6., 7. and 9., we conclude
\begin{equation} \label{eq:*_gr_z-2}
P_{n_1,0,2,0}(A) = \Span\{ y_{i_1}^{+} \cdots y_{i_s}^{+}z_{1}^{+}y_{j_1}^{+}\cdots y_{j_t}^{+}z_{2}^{+} \},
\end{equation} with $s+t = n_1$, $i_1<\dots<i_s$ and $j_1 < \dots j_t$.
In order to see that the polynomials in (\ref{eq:*_gr_z-2}) are linearly independent modulo $Id_{\mathbb{Z}_2}^{*}(A)$, suppose that \[ \sum_{I,J}\alpha_{I,J}y_{i_1}^{+} \cdots y_{i_s}^{+}z_{1}^{+}y_{j_1}^{+}\cdots y_{j_t}^{+}z_{2}^{+} = 0 \qquad \mod \Id_{\mathbb{Z}_2}^{*}(A) \]
and that there are index sets $I_0$ and $J_0$, such that $\alpha_{I_0,J_0} \neq 0$. Considering the evaluation $y_i^{+} = e_{1,1} + e_{3,3}$, $y_{j}^{+} = e_{2,2}$, $z_{l}^{+} = e_{1,2} + e_{2,3}$, for $i \in I_0$, $j \in J_0$, we obtain $\alpha_{I_0,J_0}e_{2,3} = 0$. So, $\alpha_{I_0,J_0} = 0$, a contradiction. Then, the polynomials in (\ref{eq:*_gr_z-2}) are linearly independent.
\end{enumerate}
Therefore, from items (i), (ii) and (iii), we conclude that the set of polynomials 1.-11. determines a basis for the graded $*$-identities of the algebra $A$.
\end{proof}
\subsection{Cocharacters of $(A,gr,*)$}
Let $\langle \lambda \rangle = (\lambda(1), \lambda(2), \lambda(3), \lambda(4))$ be a multipartition of $(n_1,n_2,n_3,n_4)$, i.e. $\lambda(i)\vdash n_i$. We are interested in computing the $(n_1 , \dots, n_4 )$-th cocharacter of $(A,gr,*)$,
\begin{equation}\label{eq:cocharacter_A_gr_*}
\chi_{n_1,n_2,n_3,n_4}(A) = \sum_{\langle \lambda \rangle \vdash (n_1,n_2,n_3,n_4)} m_{\langle \lambda \rangle} \chi_{\lambda(1)}\otimes \cdots \otimes \chi_{\lambda(4)}.
\end{equation}
Since $\dim (\mathcal{A}^1)^{+} = \dim (\mathcal{A}^1)^{-} = 1$, then $m_{\langle \lambda \rangle} = 0$ if $h(\lambda(i))>1$ for $i = 3$, $4$. Also, by the graded $*$-identities of $A$, we have $m_{\langle \lambda \rangle} = 0$ if $h(\lambda(1))>2$.
In the following results, we consider only the case $h(\lambda(i))\leq 1$ for $i = 3$, $4$.
First, we consider the case of even degree variables only or odd degree variables only.
\begin{proposition}
If either $\langle \lambda \rangle = ((n_1), \emptyset, \emptyset, \emptyset)$ with $n_1 >0$ or $\langle \lambda \rangle = (\emptyset, \emptyset, (n_3), (n_4))$ with $ 0 < n_3 + n_4 \leq 2$, then $m_{\langle \lambda \rangle} = 1$ in (\ref{eq:cocharacter_A_gr_*}).
\end{proposition}
\begin{proof}
Let $\langle \lambda_1 \rangle = ((n_1), \emptyset, \emptyset, \emptyset)$ and $\langle \lambda_2 \rangle = (\emptyset, \emptyset, (n_3), (n_4))$ as in the statement. Then $\omega_1 = (y_1^{+})^{n_1}$ and $\omega_2 = (z_{1}^{+})^{n_3}(z_{1}^{-})^{n_4}$ are highest weight vectors corresponding to the multipartitions $\langle \lambda_1 \rangle $ and $\langle \lambda_2 \rangle$ respectively. Since $\omega_1$ and $\omega_2$ are not polynomials identities of $A$ then $m_{\langle \lambda_i \rangle} \geq 1$. By the identities in Theorem \ref{th:Base_gr_*} we conclude that $\omega_1$ and $\omega_2$ are the only (up to a scalar) highest weight vectors corresponding to $\langle \lambda_1 \rangle $ and $\langle \lambda_2 \rangle$. Therefore, $m_{\langle \lambda_i \rangle} = 1$ for $i=1$, $2$.
\end{proof}
Before dealing with the case $n_3 + n_4 = 1$, we state a technical result (similar results can be found in \cite{cirrito2014ordinary,giambruno2019superalgebras}).
\begin{proposition} \label{pr:equation}
Modulo $\Id_{\mathbb{Z}_2}^{*}(A)$, the following equality holds:
\begin{equation} \label{eq:equation}
\begin{split}
& \underbrace{\overline{y}_{1}^{+} \cdots \widetilde{y}_{1}^{+} }_p (y_{1}^{+})^{i_1-p} z^{+} (y_{1}^{+})^{i_2-p} \underbrace{ \overline{y}_{2}^{+} \cdots \widetilde{y}_{2}^{+} }_p \\
= & \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} (y_{1}^{+})^{i_1-j} (y_{2}^{+})^{j} z^{+} (y_{1}^{+})^{i_2-p+j} (y_{2}^{+})^{p-j},
\end{split}
\end{equation}
where $i_1,$ $i_2 \geq 0$ and $p\geq 1$.
\end{proposition}
\begin{proof}
We prove it by induction on $p$.
The case $p=1$ is a straightforward computation. Let $p>1$ and let $\omega$ be the polynomial (\ref{eq:equation}).
\[
\begin{split}
\omega = & \underbrace{\overline{y}_{1}^{+} \cdots \widetilde{y}_{1}^{+} }_p (y_{1}^{+})^{i_1-p} z^{+} (y_{1}^{+})^{i_2-p} \underbrace{ \overline{y}_{2}^{+} \cdots \widetilde{y}_{2}^{+} }_p \\
= & \underbrace{\overline{y}_{1}^{+} \cdots \widetilde{y}_{1}^{+} }_{p-1} (y_{1}^{+})^{i_1-(p-1)} z^{+} (y_{1}^{+})^{i_2-1-(p-1)} {y}_{2}^{+} \underbrace{ \overline{y}_{2}^{+} \cdots \widetilde{y}_{2}^{+} }_{p-1} \\
& - \underbrace{\overline{y}_{1}^{+} \cdots \widetilde{y}_{1}^{+} }_{p-1} (y_{1}^{+})^{i_1-1-(p-1)} y_{2}^{+} z^{+} (y_{1}^{+})^{i_2-(p-1)} \underbrace{ \overline{y}_{2}^{+} \cdots \widetilde{y}_{2}^{+} }_{p-1}. \\
\end{split}
\]
Applying induction to $p-1$ we obtain
\begin{equation} \label{eq:equa-2}
\begin{split}
\omega = & \sum_{j=0}^{p-1} (-1)^{j} \binom{p-1}{j} (y_{1}^{+})^{i_1-j} (y_{2}^{+})^{j} z^{+} (y_{1}^{+})^{(i_2-1) -(p-1)+j} (y_{2}^{+})^{(p-1)-j+1} \\
& - \sum_{j=0}^{p-1} (-1)^{j} \binom{p-1}{j} (y_{1}^{+})^{(i_1-1)-j} (y_{2}^{+})^{j+1} z^{+} (y_{1}^{+})^{i_2-(p-1)+j} (y_{2}^{+})^{(p-1)-j}.
\end{split}
\end{equation}
The $j$-th monomial of the second summand is similar to the $j+1$-th monomial of the first summand, for every $j = 0$, \dots, $p-2$.
Considering the sum of these monomials, for the corresponding coefficients we have
\[
(-1)^{j+1}\binom{p-1}{j+1} - (-1)^{j}\binom{p-1}{j} = (-1)^{j+1}\left( \binom{p-1}{j+1} + \binom{p-1}{j} \right)
= (-1)^{j+1}\binom{p}{j+1}.
\]
The sum of the similar monomials of (\ref{eq:equa-2}) corresponds to the $j+1$-th monomial of
\[ \omega_1 = \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} (y_{1}^{+})^{i_1-j} (y_{2}^{+})^{j} z^{+} (y_{1}^{+})^{i_2-p+j} (y_{2}^{+})^{p-j}, \]
for $j = 0$, \dots, $p-2$. But the first monomial of the first summand of (\ref{eq:equa-2}) is equal to the first monomial of $\omega_1$, whereas the $(p-1)$-th monomial of the second summand of (\ref{eq:equa-2}) is equal to the $p$-th monomial of $\omega_1$. Thus, we have the desired equality.
\end{proof}
\begin{proposition} \label{pr:m_n3-1}
If $\langle \lambda \rangle = ((p+q,p), \emptyset, (1), \emptyset)$ or $\langle \lambda \rangle = ((p+q,p),\emptyset, \emptyset, (1))$, where $p$, $q\geq 0$, then $m_{\langle \lambda \rangle} = (q+1)$ in (\ref{eq:cocharacter_A_gr_*}).
\end{proposition}
\begin{proof}
We deal with the case $\langle \lambda \rangle = ((p+q,p), \emptyset, (1), \emptyset)$. The case $\langle \lambda \rangle = ((p+q,p),\emptyset, \emptyset, (1))$ is analogous.
Consider Young diagrams of shape $\langle \lambda \rangle$ filled in the standard way.
From the identity $[y_1^+,y_2^+]$, fixed the tableaux $ T_{\lambda(3)} =
\begin{tabular}{|c|}
\hline $ t_1 $ \\
\hline
\end{tabular}$ for an integer $t_1$, then $t_1$ must be larger than all the
integers lying in the first $p$ positions of the first row of $T_{\lambda(1)}$ and less than the ones lying in the second row. Otherwise, the corresponding highest weight vector will be a polynomial identity of $A$.
Thus, the possibilities for the standard Young tableaux, such that the corresponding highest weight vectors are linearly independent, are given by
$$T_{\lambda(1)} =
\begin{tabular}{|c|c|c|c|c|c|c|c|c|}
\hline$1$ & $2$ & $\cdots$ & $p$ & $\cdots$ & $t_1 -1$ & $t_1 + 1$ & $\cdots$ & $t_2$ \\
\hline$t_2+1$ & $t_2+2$ & $\cdots$ & $n_1$ & \multicolumn{5}{|c}{} \\
\cline { 1 - 4 }
\end{tabular} \, ,$$
$$T_{\lambda(3)} =
\begin{tabular}{|c|}
\hline $ t_1 $ \\
\hline
\end{tabular}\,, \qquad T_{\lambda(2)} = T_{\lambda(4)} = \emptyset ,$$
and the corresponding highest weight vectors are
\[ \omega_{t_1} = \overline{y}_{1}^{+} \cdots \widetilde{y}_{1}^{+} (y_{1}^{+})^{t_1-1-p} z^{+} (y_{1}^{+})^{t_2-t_1} \overline{y}_{2}^{+} \cdots \widetilde{y}_{2}^{+}, \]
where $t_1 = p+1$, \dots, $p + q +1$.
We can rewrite the polynomials $\omega_{t_1}$ as
\[ \omega_{t_1} = \overline{y}_{1}^{+} \cdots \widetilde{y}_{1}^{+} (y_{1}^{+})^{t_1-p} z^{+} (y_{1}^{+})^{t_2-p} \overline{y}_{2}^{+} \cdots \widetilde{y}_{2}^{+}, \]
where $t_1 = p, \dots, p + q $ and $t_1 + t_2 = n_1$. By Equation (\ref{eq:equation})
\[ \omega_{t_1} = \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} (y_{1}^{+})^{t_1-j} (y_{2}^{+})^{j} z^{+} (y_{1}^{+})^{t_2-p+j} (y_{2}^{+})^{p-j}.\]
Let us see that the elements of the set $\{ \omega_{t_1} \mid t_1 = p, \dots, p+q \}$ are linearly independent. Consider the evaluations $z^{+} = e_{1,2} + e_{2,3}$, $y_{i}^{+} = \alpha_i(e_{1,1} + e_{3,3}) + \beta_i e_{2,2}$. Then, \[
\begin{split}
\omega_{t_1} & = \bigg[ \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} \alpha_1^{t_1-1}\alpha_2^{j}\beta_1^{t_2-p+j}\beta_2^{p-j}\bigg] e_{1,2} + [*]e_{2,3} \\
& = \bigg[ \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} \beta_1^{j}\alpha_2^{j}\beta_2^{p-j}\alpha_1^{p-j}\alpha_1^{t_1-p}\beta_1^{t_2 -p} \bigg] e_{1,2} + [*]e_{2,3} \\
& = \big[ (\beta_2 \alpha_1 - \beta_1 \alpha_2)^{p} \alpha_1^{t_1-p}\beta_1^{t_2 -p} \big] e_{1,2} + [*]e_{2,3} .
\end{split}
\]
Here the coefficients of $e_{23}$ are irrelevant and we denote them by (*). Now, $(\beta_2 \alpha_1 - \beta_1 \alpha_2)^{p}$ appears in each evaluation of $\omega_{t_1}$, so we consider just the part $\alpha_1^{t_1-p}\beta_1^{t_2 -p}e_{1,2}$. Suppose that there are $\gamma_{i}$'s such that $ \sum_{t_1 = p}^{p+q} \gamma_{t_1}\omega_{t_1} \equiv 0$. Then, for all $\alpha_1$, $\beta_1 \in F$ we have $\sum_{t_1 = p}^{p+q} \gamma_{t_1} \alpha_1^{t_1-p}\beta_1^{t_2 -p} = 0$. It follows $\gamma_i = 0$ for each $i$. Thus, the highest weight vectors $\omega_{t_1}$ are linearly independent, $t_1 =p$, \dots, $p+q$, and $m_{\langle \lambda \rangle} = (q+1)$.
\end{proof}
Now, we consider the case $n_3 + n_4 = 2$.
\begin{proposition} \label{pr:m_n3-2}
If $\langle \lambda \rangle = ((p+q,p), \emptyset, (n_3), (n_4))$, where $p$, $q\geq 0$ and $n_3 + n_4 = 2$, then $m_{\langle \lambda \rangle} = (q+1)$ in (\ref{eq:cocharacter_A_gr_*}).
\end{proposition}
\begin{proof}
We prove the case $\langle \lambda \rangle = ((p+q,p), \emptyset, (2), \emptyset)$. The cases $\langle \lambda \rangle = ((p+q,p),\emptyset, (1), (1))$ and $\langle \lambda \rangle = ((p+q,p),\emptyset, \emptyset, (2))$ are analogous.
Consider Young diagrams of shape $\langle \lambda \rangle$ filled in the standard way.
As in the proof of Proposition \ref{pr:m_n3-1}, fixed an integer $t_1$ in the tableau $$ T_{\lambda(3)} =
\begin{tabular}{|c|c|}
\hline $ t_1 $ & $ $ \\
\hline
\end{tabular}\,,$$ $t_1$ must be larger than the integers lying in the first $p$ position of the first row of $T_{\lambda(1)}$ and less than all the ones lying in the second row. Also, by the identities $[y^{+},z_{1}z_{2}]$, $[y_{1}^{+},z_{1}y_{2}^{+}z_{2}]$, and $ z_{1}^{+}y^{+}z_{2}^{+} - z_{2}^{+}y^{+}z_{1}^{+}$, we can fix the integer $n_1+2$ in the second block of the tableaux $T_{\lambda(3)}$.
Thus, the possibilities for the standard Young tableaux, such that the corresponding highest weight vectors are linearly independent, are given by
\[
T_{\lambda(1)} =
\begin{tabular}{|c|c|c|c|c|c|c|c|c|}
\hline$1$ & $2$ & $\cdots$ & $p$ & $\cdots$ & $t_1 -1$ & $t_1 + 1$ & $\cdots$ & $t_2$ \\
\hline$t_2+1$ & $t_2+2$ & $\cdots$ & $n_1+1$ & \multicolumn{5}{|c}{} \\
\cline { 1 - 4 }
\end{tabular} \, ,
\]
\[
T_{\lambda(3)} =
\begin{tabular}{|c|c|}
\hline $ t_1 $ & $n_1+2$ \\
\hline
\end{tabular}\,, \qquad T_{\lambda(2)} = T_{\lambda(4)} = \emptyset ,
\]
and the corresponding highest weight vectors are
\[ \omega_{t_1} = \overline{y}_{1}^{+} \cdots \widetilde{y}_{1}^{+} (y_{1}^{+})^{t_1-p} z_{1}^{+} (y_{1}^{+})^{t_2-p} \overline{y}_{2}^{+} \cdots \widetilde{y}_{2}^{+}z_{1}^{+}, \]
where $t_1 = p$, \dots, $p + q $ and $t_1 + t_2 = n_1$. By Equation (\ref{eq:equation})
\[ \omega_{t_1} = \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} (y_{1}^{+})^{t_1-j} (y_{2}^{+})^{j} z_{1}^{+} (y_{1}^{+})^{t_2-p+j} (y_{2}^{+})^{p-j}z_{1}^{+}.\]
We prove the elements of the set $\{ \omega_{t_1} \mid t_1 = p, \dots, p+q \}$ are linearly independent. Consider the evaluations $z_{1}^{+} = e_{1,2} + e_{2,3}$, $y_{i}^{+} = \alpha_i(e_{1,1} + e_{3,3}) + \beta_i e_{2,2}$. Then, \[
\begin{split}
\omega_{t_1} & = \bigg[ \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} \alpha_1^{t_1-1}\alpha_2^{j}\beta_1^{t_2-p+j}\beta_2^{p-j}\bigg] e_{1,3} \\
& = \bigg[ \sum_{j=0}^{p} (-1)^{j} \binom{p}{j} \beta_1^{j}\alpha_2^{j}\beta_2^{p-j}\alpha_1^{p-j}\alpha_1^{t_1-p}\beta_1^{t_2 -p} \bigg] e_{1,3} \\
& = \big[ (\beta_2 \alpha_1 - \beta_1 \alpha_2)^{p} \alpha_1^{t_1-p}\beta_1^{t_2 -p} \big] e_{1,3} .
\end{split}
\]
Since $(\beta_2 \alpha_1 - \beta_1 \alpha_2)^{p}$ appears in each evaluation of $\omega_{t_1}$, we consider only the part $\alpha_1^{t_1-p}\beta_1^{t_2 -p}e_{1,2}$. Suppose there exist $\gamma_{i}$'s such that $\sum_{t_1 = p}^{p+q} \gamma_{t_1}\omega_{t_1} \equiv 0$. Then, for all $\alpha_1$, $\beta_1 \in F$ one has $\sum_{t_1 = p}^{p+q} \gamma_{t_1} \alpha_1^{t_1-p}\beta_1^{t_2 -p} = 0$. It follows that $\gamma_i = 0$ for each $i$. So the highest weight vectors $\omega_{t_1}$ are linearly independent for $t_1 =p$, \dots, $p+q$ and $m_{\langle \lambda \rangle} = (q+1)$.
\end{proof}
Hence we obtain the following theorem.
\begin{theorem}
Let
\[ \chi_{n_1,n_2,n_3,n_4}(A) = \sum_{\langle \lambda \rangle \vdash (n_1,n_2,n_3,n_4) } m_{\langle \lambda \rangle}\chi_{\lambda(1)} \otimes \chi_{\lambda(2)} \otimes \chi_{\lambda(3)} \otimes \chi_{\lambda(4)} \]
be the $(n_1,n_2,n_3,n_4)$-cocharacter of $A$. Then
\begin{itemize}
\item[(1)] $m_{\langle \lambda \rangle} = 1$, if $\langle \lambda \rangle = ((n_1), \emptyset, \emptyset, \emptyset) $ or $\langle \lambda \rangle = (\emptyset, \emptyset, (n_3),(n_4))$, where $n_1 > 0$ and $0< n_3 + n_4 \leq 2$.
\item[(2)] $m_{\langle \lambda \rangle} = q+1$, if $\lambda = ((p+q,p), \emptyset, (n_3), (n_4))$, where $p,q \geq 0$ and $n_3 + n_4 = 1$;
\item[(3)] $m_{\langle \lambda \rangle} = q+1$, if $\lambda = ((p+q,p), \emptyset, (n_3), (n_4))$, where $p,q \geq 0$ and $n_3 + n_4 = 2$.
\end{itemize}
In all remaining cases $m_{\langle \lambda \rangle} = 0$.
\end{theorem}
\bibliographystyle{amsplain}
\bibliography{references}
\end{document}
\providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}
\providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }
% \MRhref is called by the amsart/book/proc definition of \MR.
\providecommand{\MRhref}[2]{%
\href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}
}
\providecommand{\href}[2]{#2}
\begin{thebibliography}{10}
\bibitem{cirrito2013gradings}
A.~Cirrito, \emph{Gradings on the algebra of upper triangular matrices of size
three}, Linear Algebra and its Applications \textbf{438} (2013), no.~11,
4520--4538.
\bibitem{cirrito2014ordinary}
A.~Cirrito and F.~Martino, \emph{Ordinary and graded cocharacter of the
{J}ordan algebra of $2\times 2$ upper triangular matrices}, Linear Algebra
and its Applications \textbf{451} (2014), 246--259.
\bibitem{di2006involutions}
O.~M. {D}i Vincenzo, P.~Koshlukov, and R.~{L}a Scala, \emph{Involutions for
upper triangular matrix algebras}, Advances in Applied Mathematics
\textbf{37} (2006), no.~4, 541--568.
\bibitem{di2004}
O.~M. {D}i Vincenzo, P.~Koshlukov, and A.~Valenti, \emph{Gradings on the
algebra of upper triangular matrices and their graded identities}, Journal of
Algebra \textbf{275} (2004), no.~2, 550--566.
\bibitem{drensky2000free}
V.~Drensky, \emph{Free algebras and {PI}-algebras: Graduate course in algebra},
Springer, Singapore, 2000.
\bibitem{giambruno2019superalgebras}
A.~Giambruno, A.~Ioppolo, and D.~{L}a Mattina, \emph{Superalgebras with
involution or superinvolution and almost polynomial growth of the
codimensions}, Algebras and Representation Theory \textbf{22} (2019), no.~4,
961--976.
\bibitem{gordienko2009regev}
A.~S. Gordienko, \emph{Regev’s conjecture and codimensions of {PI} algebras},
Acta applicandae mathematicae \textbf{108} (2009), no.~1, 33--55.
\bibitem{pktcm}
P.~Koshlukov and T.~C.~de Mello, \emph{The centre of generic algebras of small
{P}{I} algebras}, Journal of Algebra \textbf{375} (2013), 109--120.
\bibitem{koshlukov2013polynomial}
\bysame, \emph{On the polynomial identities of the algebra {$M_{1,1}(E)$}},
Linear Algebra and its Applications \textbf{438} (2013), no.~11, 4469--4482.
\bibitem{MISHCHENKO200066}
S.~Mishchenko and A.~Valenti, \emph{A star-variety with almost polynomial
growth}, Journal of Algebra \textbf{223} (2000), no.~1, 66--84.
\bibitem{popov}
Angel~P. popov, \emph{Identities of the tensor square of a grassmann algebra},
Algebra Logic \textbf{21} (1982), no.~4, 442--471.
\bibitem{valenti2002graded}
A.~Valenti, \emph{The graded identities of upper triangular matrices of size
two}, Journal of Pure and Applied Algebra \textbf{172} (2002), no.~2-3,
325--335.
\bibitem{valenti2007group}
A.~Valenti and M.~Zaicev, \emph{Group gradings on upper triangular matrices},
Archiv der Mathematik \textbf{89} (2007), no.~1, 33--40.
\bibitem{valenti2009graded}
\bysame, \emph{Graded involutions on upper-triangular matrix algebras}, Algebra
Colloquium, vol.~16, World Scientific, 2009, pp.~103--108.
\end{thebibliography}