Chapter 23 Representation of a Scheme

Monday, March 22, 1993

Theorem 23.1 Let \(Y = (X, \{R_i\}_{0\leq i\leq D})\) be a symmetric scheme. (View the standard module \(V\) as an algebra of functions from \(X\) to \(\mathbb{C}\).) Then the following are equivalent.

\((i)\) \(Y\) is \(Q\)-polynomial with respect to ordering \(E_0, E_1, \ldots, E_D\) of primitive idempotents.
\((ii)\) For all \(i\) \(\:(0\leq i\leq D)\),

\[E_0V + E_1V + (E_1V)^2 + \cdots + (E_1V)^i = E_0V + E_1V + \cdots + E_iV.\]

Proof.

By Lemma 20.4 \((ii)\), \((iii)\).

\[E_h(E_iV\circ E_jV) = 0 \text{ if and only if } q^h_{ij} = 0 \quad (0\leq i,j,h\leq D).\]

\((i)\to(ii)\) By our assumption,

\[q^h_{1j} = 0 \text{ if } |h-j|>1, \text{ and } q^{j+1}_{1j}\neq 0.\] So, \[\begin{equation} E_1V\circ E_jV \subseteq E_{j-1}V + E_jV + E_{j+1}V \quad (0\leq j\leq D), \tag{23.1} \end{equation}\] \[\begin{equation} E_{j+1}(E_1V\circ E_jV) = E_{j+1}V \quad (0\leq j\leq D-1), \tag{23.2} \end{equation}\] by Lemma 20.4.

Also \(E_0V \subseteq \mathrm{Span}(\delta)\), where \(\delta\) is all 1’s vector, i.e., \(1\) as a function \(X\to \mathbb{C}\). So, \[\begin{equation} E_0V\circ E_jV = E_jV \quad (0\leq j\leq D). \tag{23.3} \end{equation}\] Show \((ii)\) by induction on \(i\).

The cases \(i=0, 1\) are trivial.

\(i>1\): \(\subseteq\). \[\begin{align} & E_0V + E_1V + (E_1V)^2 + \cdots + (E_1V)^i\\ & \quad = E_0V + E_1V\circ (E_0V + E_1V + \cdots + (E_1V)^{i-1})\\ & \quad = E_0V + E_1V\circ (E_0V + E_1V + \cdots + E_{i-1}V)\\ & \quad \subseteq E_0V + E_1V + \cdots + E_{i}V \end{align}\] by (23.1).

\(\supseteq\).

Claim. \(E_iV\subseteq E_1V\circ E_{i-1}V + E_{i-1}V + E_{i-2}V \quad (2\leq i\leq D)\).

Proof of Claim. By (23.2), \[E_i(E_1V \circ E_{i-1}V) = E_iV.\] For all \(v\in E_i V\), there exists \(u\in E_1V\circ E_{i-1}V\) such that \(E_iu = v\).

On the other hand, by (23.1), \[E_1V\circ E_{i-1}V \subseteq E_{i-2}V + E_{i-1}V + E_{i-2}V.\] So, \(u = w+v\), where \(w\in E_{i-2}V + E_{i-1}V\). We have, \[w = u-v \in E_1V \circ E_{i-1}V + E_{i-1}V + E_{i-2}V\] as desired.

HS MEMO

\[E_iV \circ E_jV = \mathrm{Span}(u\circ v \mid u\in E_iV, v\in E_jV).\]

By claim, \[\begin{align} & E_0V + E_1V + \cdots + E_iV\\ & \quad \subseteq E_0V + E_1V + \cdots + E_{i-1}V + E_1V\circ E_{i-1}V\\ & \quad \subseteq E_0V + E_1V + \cdots + (E_{1}V)^{i-1} + E_1V(E_0V + E_1V + \cdots + (E_{1}V)^{i-1})\\ & \quad \subseteq E_0V + E_1V + \cdots + (E_{1}V)^{i-1} + (E_1V)^{i}. \end{align}\]

\((ii)\to(i)\)

Claim 1. Pick \(i, j\) \((0\leq i,j\leq D)\) with \(j>i+1\). Then \(q^j_{1i} = 0\).

Proof of Claim 1. \[\begin{align} E_j(E_1\circ E_jV) & \subseteq E_j(E_1V\circ(E_0V + E_1V + (E_1V)^2 + \cdots + (E_1V)^i))\\ & \subseteq E_j(E_0V + E_1V + (E_1V)^2 + \cdots + (E_1V)^{i+1})\\ & = E_j(E_0V + E_1V + \cdots + E_{i+1}V)\\ & = 0. \end{align}\] So \(q^j_{1i}=0\) by Lemma 20.4.

Claim 2. \(q^{i+1}_{1i} \neq 0\) \((0\leq i < D)\).

Proof of Claim 2. \[\begin{align} & E_0V + E_1V + \cdots + E_{i+1}V\\ & \quad = E_0V + E_1V + \cdots + (E_1V)^{i+1}\\ & \quad = E_0V + E_1V\circ(E_0V + E_1V + \cdots + (E_{1}V)^{i})\\ & \quad = E_0V + E_1V\circ(E_0V + E_1V + \cdots + E_iV)\\ & \quad = E_0V + E_1V\circ(E_0V + \cdots + E_iV). \end{align}\] So, \[\begin{align} E_{i+1}V & = E_{i+1}(E_1V\circ (E_0V + \cdots + E_iV))\\ & = E_{i+1}(E_1V \circ E_iV) \end{align}\] by Claim 1 and Lemma 20.4.

Hence, \(q^{i+1}_{1i}\neq 0\) by Lemma 20.4.

Let \(Y = (X, \{R_i\}_{0\leq i\leq D})\) be a commutative scheme with standard module \(V\).

Definition 23.1 A representation of \(Y\) is a pair \((\rho, H)\), where \(H\) is a non-zero Hermitean space (with inner product \(\langle \;, \;\rangle\)) and \(\rho: X\to H\) is a map satisfying the following.

\(\mathrm{R1}\). \(H = \mathrm{Span}(\rho(x)\mid x\in X)\).
\(\mathrm{R2}\). \(\langle \rho(x), \rho(y)\rangle\) depends only on \(i\) for which \((x,y)\in R_i\) \((x,y\in X)\).
\(\mathrm{R3}\). For every \(x\in X\) and for all \(i\) \((0\leq i\leq D)\),

\[\sum_{y\in X, (y,x)\in R_i}\rho(y)\in \mathrm{Span}(\rho(x)).\]

Above representation is nondegenerate if \(\{\rho(x)\mid x\in X\}\) are distinct.

Example 23.1 \(Y = H(D,2)\), \(X = \{a_1\cdots a_D\mid a_i\in \{1,-1\}, 1\leq i\leq D\}\). Let \(H = \mathbb{C}^D\) and \(\langle \;, \;\rangle\) usual Hermitean dot product.

For a vertex \(x = a_1\cdots a_D\in X\), define \[\rho(x) = a_1\cdots a_D\in H.\] Then, \(\mathrm{R1}-\mathrm{R3}\) hold.

HS MEMO

\(\mathrm{R1}, \mathrm{R2}\) are obvious. For \(\mathrm{R3}\), we may assume that \(x = 1\cdots 1\). Restrict \[\sum_{y\in X, (y,x)\in R_i}\rho(y)\] on the first coordinate. Then, \[\begin{align} -1 & \quad \text{appears }\; \binom{D-1}{i-1} \;\text{ times}\\ 1 & \quad \text{appears } \;\binom{D-1}{i} \;\text{ times}. \end{align}\] Hence, \[\sum_{y\in X, (y,x)\in R_i}\rho(y) = \left(\binom{D-1}{i} - \binom{D-1}{i-1}\right)\rho(x).\]

Let \((\rho, H)\) be a representation of arbitrary commutative scheme \(Y\). Set \[E = (\langle \rho(x),\rho(y)\rangle)_{x,y\in X}\] Gram matrix of the representation.

Definition 23.2 Representations \((\rho, H)\), \((\rho', H')\) of \(Y\) are equivalent, whenever, Gram matrices are related by \[E'\in \mathrm{Span} E.\] We do not distinguish between equivalent representations.

Note. Suppose \((\rho, H)\) is a representation of a symmetric scheme \(Y\). Pick \(x,y\in X\) with \((x,y)\in R_j\).

Then \((y,x)\in R_j\). So, by \(\mathrm{R2}\), \[\langle \rho(x), \rho(y)\rangle = \langle \rho(y),\rho(x)\rangle = \overline{\langle \rho(x), \rho(y)\rangle},\] since \(\langle \;, \;\rangle\) is Hermitean.

Hence, the Gram matrix \(E\) of \(\rho\) is real symmetirc. Without loss of generality, we can view \(H\) as a real Euclidean space in this case.

Lemma 23.1 Let \(Y = (X, \{R_i\}_{0\leq i\leq D})\) be a commutative scheme and \(V\) a standard module.

Let \(E_j\) be any primitive idempotent of \(Y\).

\((i)\) \((\rho, H)\) is a representation of \(Y\), where \(H = E_jV\) (with inner product inherited from \(Y\)).

\[\rho: X \to H \quad (x\mapsto E_j\hat{x})\] (i.e., \(\rho(x)\) is the \(x\)-th column of \(E_j\).)

\((ii)\) \(\langle \rho(x),\rho(y)\rangle = |X|^{-1}q_j(i)\), if \((x,y)\in R_i\), \((x,y\in X)\).
\((iii)\) For \(0\leq i\leq D\) and \(x,y\in X\),

\[\sum_{y\in X, (y,x)\in R_i}\rho(y) = p_i(j)\rho(x).\]

\((iv)\) \((\rho,H)\) is nondegenerate if and only if \(q_j(i) \neq q_j(0)\) for all \(i\), \((0\leq i\leq D)\).
\((v)\) Every representation of \(Y\) is equivalent to a representation of the above type for some \(j\) \((0\leq j\leq D)\), and \(j\) is unique.

Proof.

\((i)-(iii)\).

\(\mathrm{R1}\): \(\mathrm{Span}(\rho X)\) is the column space of \(E_j\) which is equal to \(H\).

\(\mathrm{R2}\): \[\begin{align} \langle \rho(x),\rho(y)\rangle & = \langle E_j\hat{x}, E_j\hat{y}\rangle \\ & = (\overline{E_j\hat{x}})^\top E_j\hat{y}\\ & = \hat{x}^\top \overline{E_j}^\top E_j\hat{y}\\ & = \hat{x}^\top E_j \hat{y}\\ & = (E_j)_{xy}. \end{align}\] Note that \(\overline{E_j}^\top = E_j\) by Lemma 19.1.

Recall \[E_j = |X|^{-1}(q_j(0)A_0 + \cdots + q_j(D)A_D).\] So, \[(E_j)_{xy} = |X|^{-1}q_j(i), \quad \text{ where } \; (x,y)\in R_i.\]

\(\mathrm{R3}\): Recall \[A_i = p_i(0)E_0 + \cdots + p_i(D)E_D.\] So, \(E_jA_i = p_i(j)E_j\), and \[p_i(j)\rho(x) = p_i(j)E_j\hat{x} = E_jA_i\hat{x} = E_j\sum_{y\in X, (y,x)\in R_i}\hat{y} = \sum_{y\in X, (y,x)\in R_i}\rho(y).\]

Note. \[A_i\hat{x} = \sum_{y\in X, (x,y)\in R_{i'}}\hat{y}.\]

Pf. \[\begin{align} \text{$z$ entry of LHS} & = (A_i\hat{x})_z \\ & = \sum_{w\in X}(A_i)_{zw}\hat{x}_w\\ & = (A_i)_{zx}\\ & = \begin{cases} 1 & \text{if $(x,z)\in R_{i'}$}\\ 0 & \text{else}. \end{cases} \end{align}\] \[\begin{align} \text{$z$ entry of RHS} & = \sum_{y\in X, (x,y)\in R_{i'}, z = y}1\\ & = \begin{cases} 1 & \text{if $(x,z)\in R_{i'}$}\\ 0 & \text{else}. \end{cases} \end{align}\]

\((iv)\) By \((ii)\),

\[\begin{align} \|\rho(x)\|^2 & = \langle \rho(x), \rho(y)\rangle\\ & = |X|^{-1}q_j(0)\\ & = |X|^{-1}m_j, \end{align}\] as \(m_j = \dim E_jV\), and is independent of \(x\in X\).

Pick distinct \(x,y\in X\) such that \((x,y)\in R_i\) with \(i\neq 0\).

Then, \[\begin{align} \rho(x) = \rho(y) & \Leftrightarrow \langle \rho(x),\rho(y)\rangle = \|\rho(x)\|^2 = |X|^{-1}q_j(0)\\ & \Leftrightarrow |X|^{-1}q_j(i) = |X|^{-1}q_j(0)\\ & \Leftrightarrow q_j(i) = q_j(0). \end{align}\]

Hence, we have \((iv)\). To be continued.