Chapter 21 Norton Algebras
Wednesday, March 17, 1993
Proof (Proof of Lemma 20.4).
\[v = \sum_{x\in X}\alpha_x \hat{x}.\] Pick a vertex \(z\in X\) and compare \(z\)-coordinate of each side in \((i)\). \[\begin{align} (A^*_j(x)v)_z & = (A^*_j(x))_{zz}v_z = |X|(E_j)_{xz}\alpha_z.\\ |X|(E_{\hat{j}}\hat{x}\circ v)_z & = |X|(E_{\hat{j}}\hat{x})_z\cdot \alpha_z = |X|(E_j)_{xz}\alpha_z. \end{align}\] Note that \(E_{\hat{j}}\hat{x}\) is the column \(x\) of \(E_{\hat{j}}\), which is the row \(x\) of \(E_j\).
Claim. \(E_h(E_iV \circ E_jV) = 0\).
\[\begin{align} E_h(E_iV \circ E_jV) & = E_h(\mathrm{Span}(v\circ w\mid v\in E_iV, w\in E_jV))\\ & = E_h(\mathrm{Span}(E_i\hat{y}\circ E_j\hat{z}\mid y,z\in X))\\ & = \mathrm{Span}(E_h(E_j\hat{z}\circ E_i\hat{y}\mid y,z\in X)\\ & = \mathrm{Span}((E_hA^*_{\hat{j}}(z)E_i)\hat{y}\mid y,z\in X) && \text{by $(i)$.} \end{align}\] But \(q^h_{ij} = 0\) implies \(q^{\hat{h}}_{\hat{j}\hat{i}} = 0\).
So, by Lemma 20.3 \((ii)\), \[ 0 = (E_{\hat{i}}A^*_{\hat{j}}E_{\hat{h}})^\top = E_h A^*_{\hat{j}}E_i.\] Hence, \(E_h(E_iV\circ E_jV) = 0\).
\[E_h(E_iV \circ E_jV)\subseteq E_hV\] is clear. We show the other inclusion. Since \[\begin{align} E_i\hat{y} \circ E_j\hat{y} &= (\text{column $y$ of $E_i$}\circ \text{column $y$ of $E_j$}) \\ & = \text{column $y$ of $E_i\circ E_j$}\\ & = (E_i\circ E_j)\hat{y}\\ & = \left(\frac{1}{|X|}\sum_{h=0}^D q^h_{ij}E_h\right)\hat{y}, \end{align}\] we have, \[\begin{align} E_h(E_iV\circ E_jV) & = E_h\mathrm{Span}(E_i\hat{y}\circ E_j\hat{z}\mid y,z\in X)\\ & \supseteq E_h \mathrm{Span}(E_i\hat{y}\circ E_j\hat{y}\mid y\in X)\\ & = \mathrm{Span}(q^h_{ij}E_h\hat{y}\mid y\in X)\\ & = \mathrm{Span}(E_h\hat{y}\mid y\in X) && \text{since $q^h_{ij}\neq 0$}\\ & = E_hV. \end{align}\] This proves the assertion.
Lemma 21.1 Given a commutative scheme \(Y = (X, \{R_i\}_{0\leq i\leq D})\), fix \(j\) \((0\leq j\leq D)\). Define a binary multiplication: \[E_jV \times E_jV \longrightarrow E_jV \quad ((v,w) \mapsto v\ast w = E_j(v\circ w)).\] Then,
In particular, the vector space \(E_jV\) together with \(\ast\) is a commutative \(\mathbb{C}\)-algebra, (not associative in general).
(\(N_j: (E_jV, \ast)\) is called the Norton algebra on \(E_jV\).)
Let \(Y\), \(j\), \(N_j\) be as in Lemma 21.1, and \(M\) Bose-Mesner algebra of \(Y\). Let \[\begin{align} \mathrm{Aut}Y & = \{\sigma\in \mathrm{Mat}_X(\mathbb{C}) \mid \sigma: \text{ permutation matrix }, \sigma \cdot m = m\cdot \sigma \;\text{ for all }m\in M\}\\ & = \{\sigma\in \mathrm{Mat}_X(\mathbb{C}) \mid \sigma: \text{ permutation matrix },\\ & \qquad (x,y)\in R_i \to (\sigma x, \sigma y)\in R_i, \text{ for all } i, \text{ and for all } x,y\in X\}\\ \mathrm{Aut}(N_j) & = \{\sigma: E_jV \to E_jV \mid \sigma \text{ is a $\mathbb{C}$-algebra isomorphim},i.e.,\\ & \qquad \sigma(v\ast w) = \sigma(v)\ast\sigma(w) \text{ for all }v, w\in E_jV\}. \end{align}\]
Lemma 21.2 Let \(Y, j, \ast\) be as in Lemma 21.1.
(i.e., a representation of \(\mathrm{Aut}(Y)\)).
Proof.
\[\sigma E_j v = E_j\sigma v,\] since \(\sigma\) commutes with each element of \(M\).
\[\sigma(v\ast w) = \sigma(E_j(E_jv\circ E_jw)) = E_j\sigma(E_jv\circ E_jw) = E_j(E_j\sigma v\circ E_j\sigma w) = \sigma(v)\ast \sigma(w).\]
\[M = \{m\in \mathrm{Mat}_X(\mathbb{C})\mid \sigma\cdot m = m\cdot \sigma, \text{ for all }\sigma\in \mathrm{Aut}(Y)\}.\] Suppose there sia a nonzero proper subspace \(0\neq W\subsetneq E_jV\) that is \(\mathrm{Aut}(Y)\)-invariant.
Set \[W^\bot = \{v\in E_jV\mid \langle w, v\rangle = 0, \text{ for all }w\in W\}.\] Then, \(W^\bot\) is a module for \(\mathrm{Aut}(Y)\), since \(\mathrm{Aut}(Y)\) is closed under transpose conjugate.
Let \(e: V\to W\) and \(f: V\to W^\bot\) be the orthogonal projection such that \(e + f = E_j\), \[e^2 = e, f^2 = f, ef = fe = 0, eE_h = 0, \text{ if } h\neq j.\]
Since \(e\) commutes with all \(\sigma\in \mathrm{Aut}(Y)\), \(e\in M\) and \[e = \sum_{i=0}^D \alpha_i E_i.\] If \(h\neq j\), then \(0 = eE_h\) and \(\alpha_h = 0\). Thus, \(e = \alpha_jE_j\), i.e., \(e=0\) or \(f=0\).
A contradiction.
Norton algebras were used in original construction of Monster, a finite simple group \(G\).
Compute character table of \(G\),
\(\quad\to\) \(p^h_{ij}\), \(q^h_{ij}\) of group scheme on \(G\),
\(\quad\to\) find \(j\) where \(m_j = \dim E_jV\) is small and \(q^j_{jj}\neq 0\),
\(\quad\to\) guess abstract structure of \(N_j\) using the knowlege of \(p^h_{ij}\)’s and \(q^h_{ij}\)’s,
\(\quad\to\) compute \(\mathrm{Aut}(N_j)\),
\(\quad\to\) \(G\).