Definition 1
7A0SDefine \text{Aff}^{1}(S_{\ell}) to be the unital associative algebra over \mathbb{C} with generators \sigma^{\pm 1}_i, i \in [1, \ell-1 ], and \mathbf{y}_p, p \in [ 1 , \ell ] with relations
for all applicable m,p, k,j.
We state and prove an analog of the Schur-Weyl duality for a quotient of the classical 2-toroidal Lie algebra of type A. We then provide a method to extend this duality to the m-toroidal case, m > 2.
Define \text{Aff}^{1}(S_{\ell}) to be the unital associative algebra over \mathbb{C} with generators \sigma^{\pm 1}_i, i \in [1, \ell-1 ], and \mathbf{y}_p, p \in [ 1 , \ell ] with relations
for all applicable m,p, k,j.
Equip M \otimes_{\mathbb{C}[S_{\ell}]} V^{\otimes \ell} with the natural left U(\mathfrak{sl}_{n+1})–module structure induced by that on V^{\otimes \ell} (that is, x.(m \otimes \mathbf{w}) = m \otimes x.\mathbf{w} for m \in M, \mathbf{w} \in V^{\otimes \ell} and x\in\mathfrak{sl}_{n+1}). Then, if \ell \leq n, the functor M \to M \otimes_{\mathbb{C}[S_{\ell}]} V^{\otimes \ell} is an equivalence from the category of right finite dimensional S_{\ell}–modules to the category of left finite dimensional U(\mathfrak{sl}_{n+1})–modules of degree \ell.
Fix integers \ell \geq 0, n \geq 1. There exists a functor \mathcal{F} from the category \text{Rep } \text{Aff}^{1}(S_{\ell}) of finite dimensional right \text{Aff}^{1}(S_{\ell})–modules to the category \text{Rep}_{\ell} (\mathfrak{\tilde{sl}}_{n+1}) of finite dimensional level 0 left \mathfrak{\tilde{sl}}_{n+1}–modules of degree \ell defined as follows. If M is an \text{Aff}^{1}(S_{\ell})–module, then the \mathfrak{sl}_{n+1}–module structure on \mathcal{F}(M) = M \otimes_{\mathbb{C}[S_{\ell}]} V^{\otimes \ell} extends to an \mathfrak{\tilde{sl}}_{n+1}–module structure by
where m \in M and \mathbf{v} \in V^{\otimes \ell}. Further, \mathcal{F} is an equivalence of categories if \ell \leq n.
[Flicker] states that \mathcal{F} is not an equivalence when \ell = n; however, our n is n-1 in [Flicker]. We have adjusted the statement of the theorem accordingly. Note also the alternative statement [Flicker] Theorem 1.2 which makes explicit the action of an arbitrary element of L(\mathfrak{sl}_{n+1}).
Define \operatorname{Aff}^2(S_{\ell}) to be the unital associative algebra over \mathbb{C} with generators \sigma^{\pm 1}_i, i \in [1, \ell-1 ], \mathbf{x}_r, \mathbf{y}_r, r \in [1, \ell ] and relations
for k \in [1, \ell-2 ] in the second relation; k,j \in [1, \ell-1 ] otherwise; and m,p \in [1, \ell ].
The subalgebra generated by \sigma_i, \mathbf{y}_j is isomorphic to \text{Aff}^{1}(S_{\ell}) \cong \mathbb{C} [\mathbb{Z}^{\ell} \rtimes S_{\ell}] from Definition 1, and the same is true for the subalgebra generated by \sigma_i, \mathbf{x}_j. Further, if \ell=1 in Definition 4, most relations above would be trivial, and the algebra \mathbb{C}[\mathbf{x}_1^{\pm 1}, \mathbf{y}_1^{\pm 1}] would be the coordinate algebra of the algebraic torus \left( \mathbb{C}^* \right)^2.
Definition 4 specializes that of the toroidal Hecke algebra given in [VV96] Definition 1.1. From [VV96] Remark 1.4, the toroidal Hecke algebra takes the affine Hecke algebra (Def. 3.1 in [Chari]) and adds a second set of polynomial generators. In this article, starting from [VV96] Definition 1.1, we take the classical limit q \to 1 and take \mathbb{C} in place of \mathbb{C}[s^{\pm 1}, t^{\pm 1}] (that is, we set s=1=t).
We have \mathbf{x}_m\mathbf{y}_p= \mathbf{y}_p \mathbf{x}_m for all m,p\in[1,\ell].
If \ell=1, then the equality is just a defining relation. If \ell=2 then we use the defining relations as follows:
and further use of the defining relations again give the proposition for all m,p. If \ell >2 we proceed by a double induction. Observe
and an induction proves the result for all p\ge 2. It follows that \mathbf{x}_1 \mathbf{y}_1\cdots \mathbf{y}_\ell= \mathbf{y}_1 \mathbf{x}_1 \mathbf{y}_2\cdots \mathbf{y}_\ell which implies that \mathbf{x}_1\mathbf{y}_1= \mathbf{x}_1\mathbf{y}_1.
Assuming now that the result holds for all 1\le j<m we prove it for m. Here we find that
and hence it suffices to prove that \mathbf{x}_m\mathbf{y}_{m+1}= \mathbf{y}_{m+1}\mathbf{x}_m and \mathbf{x}_m\mathbf{y}_m= \mathbf{y}_m\mathbf{x}_m. To prove the first we use the inductive hypothesis on m and then the defining relations to get
The second equality follows by using the inductive hypothesis to see that \sigma_{m-1}\mathbf{y}_{m-1}\mathbf{x}_{m-1} \sigma_{m-1}^{-1}=\sigma_{m-1}\mathbf{x}_{m-1}\mathbf{y}_{m-1}\sigma_{m-1}^{-1} and then using the defining relations.
\operatorname{Aff}^2(S_{\ell}) \cong \mathbb{C} [(\mathbb{Z}^{\ell} \times \mathbb{Z}^{\ell}) \rtimes S_{\ell}] = \mathbb{C} [\mathbb{Z}^{\ell} \rtimes (\mathbb{Z}^{\ell} \rtimes S_{\ell})].
Suppose that M is a finite dimensional right module for \operatorname{Aff}^2(S_{\ell}) and set
The following formulae define an action of \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}] on \mathcal F(M): for i\in[0,n] and k\in\mathbb Z,
for all m\in M and \mathbf{v} \in V^{\otimes \ell}. Moreover, if M' is another finite dimensional module for \operatorname{Aff}^2(S_{\ell}) and f: M\to M' is a map of \operatorname{Aff}^2(S_{\ell})–modules, the assignment m\otimes \mathbf{v}\to f(m)\otimes \mathbf{v} extends to a homomorphism of \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]–modules.
A direct verification using the relations in \operatorname{Aff}^2(S_{\ell}) shows that \mathcal{F} is well-defined on objects (that is, x.(m.\sigma_i \otimes_{\mathbb{C}[S_{\ell}]} \mathbf{v}) = x.(m \otimes_{\mathbb{C}[S_{\ell}]} \sigma_i.\mathbf{v}) for each i, with x a generator of \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}], and m \in M, \mathbf{v} \in V^{\otimes \ell}) as in [Flicker] §5. It is elementary to see that \mathcal{F} is well-defined on morphisms, a homomorphism of \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]–modules, and that \mathcal{F}(\textbf{1}_{\text{Rep}\left(\operatorname{Aff}^2(S_{\ell})\right)}) = \textbf{1}_{\text{Rep}_{\ell}\left(\mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]\right)} and \mathcal{F}(f \circ g) = \mathcal{F}(f) \circ \mathcal{F} (g). Thus \mathcal{F} exists and is a well-defined, covariant functor. To check that the formulae give an action of \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}] it suffices to prove that the relations in equations (?)-(?) are satisfied. But this is a straightforward, if tedious, computation. The final statement of the proposition is also obvious.
Retain the notation established so far. There exists a covariant functor \mathcal{F}: \text{Rep}\left(\operatorname{Aff}^2(S_{\ell})\right) \to \text{Rep}_{\ell}(\mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]) which sends M\to\mathcal F(M) and the morphism f:M\to M' to the morphism \mathcal{F}(f). Further, \mathcal{F} is an equivalence of categories when \ell \leq n.
Recall, by [Moody] Proposition 3.5,
generate \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]. Taking k=0 reduces Theorem 7 to Theorem 3. Theorem 7 therefore extends the statement for \mathfrak{\tilde{sl}}_{n+1} to \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]. Note also that \mathbf{x} (resp. \mathbf{y}) corresponds to the variable s (resp. t).
The functor \mathcal{F}: \text{Rep}\left(\operatorname{Aff}^2(S_{\ell})\right) \to \text{Rep}_{\ell}\left(\mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]\right) is fully faithful.
The classical Schur-Weyl duality gives us that the functor
is fully faithful. Since any morphism \phi: M\to N of \operatorname{Aff}^2(S_{\ell})–modules is also a morphism of S_{\ell}–modules and \mathcal F(\phi)=\mathcal F^0(\phi) the injectivity of \mathcal F is immediate.
For surjectivity, let \tilde\phi: M \otimes_{\mathbb{C}[S_{\ell}]} V^{\otimes \ell} \rightarrow N \otimes_{\mathbb{C}[S_{\ell}]} V^{\otimes \ell} be a map of \mathfrak{sl}_{n+1}[t^{\pm 1}, s^{\pm 1}]–modules and hence a map of \mathfrak{sl}_{n+1}–modules. Using the classical Schur–Weyl duality we can choose an S_{\ell}–module map \phi:M\to N
such that \mathcal F^0(\phi)=\tilde\phi. It suffices to check that \phi is a map of \operatorname{Aff}^2(S_{\ell})–modules and for this, we must check that \phi commutes with the generators \mathbf{x}_j, \mathbf{y}_j for j\in[1,\ell].
By considering h_i(k) commuting with \tilde\phi (using the classical Schur-Weyl duality), then \tilde\phi commutes with \mathbf{x}_p^k
action on m \otimes v_1 \otimes \cdots \otimes v_\ell.
An argument identical to the one given in [Flicker], (using the action of x_0^+(0) acting on m \otimes_{\mathbb{C}[S_{\ell}]} v_1 \otimes \cdots \otimes v_\ell for suitable choices of v_j, 1\le j\le \ell), we get that \phi commutes with \mathbf{y}_1.
To see that \phi commutes with \mathbf{y}_r for r>1 assume that we have proved the result for r-1 and use
where we have used the inductive hypothesis for the second equality.
Suppose that i_1,\ldots, i_\ell\in[1,n+1] are distinct and set \mathbf{v} = v_{i_1} \otimes \cdots \otimes v_{i_{\ell}} \in V^{\otimes \ell}. Then V^{\otimes \ell} = U(\mathfrak{sl}_{n+1}).\mathbf{v} and the map M\to M\otimes_{\mathbb{C}[S_\ell]}V^{\otimes\ell} defined by m \mapsto m \otimes\mathbf{v} is injective for any S_{\ell}–module M.
For j_\ell\ne i_\ell we have
It follows that
Repeating with x_{j_{r}, i_{r}} for 1\le r\le \ell-1 , the first assertion of the lemma follows. The injectivity is now immediate since m\otimes \mathbf{v}=0 would imply that m\otimes V^{\otimes \ell}=0.
For each i\in [0, n] , k \in \mathbb{Z}, there exist \alpha_{j,x_i^\pm(k)},\alpha_{j,h_i(k)} \in \text{End}_{\mathbb{C}}(M) with
and moreover,
For a fixed i, the set \{ \tau.u^{\pm}_{i,1} \mid \tau \in S_{\ell} \} spans the \mathfrak{sl}_{n+1}-weight space V^{\otimes \ell}_{\lambda_{u^{\pm}_{i,1}}} of weight \lambda_{u^{\pm}_{i,1}} of V^{\otimes \ell} where
and \lambda_{u^{-}_{i,1}}=\lambda_{u^{+}_{i,1}}-\alpha_i. Indeed, using the action from § ?, for any h_i (0), i = 1, \ldots , n, we have
Thus v_j is of weight \varepsilon_j and hence h_j (k).\tau.u^{\pm}_{i,1} = \lambda_{u^{\pm}_{i,1}} (h_j) \tau.u^{\pm}_{i,1} and so \tau. u^{\pm}_{i,1} is of \mathfrak{sl}_{n+1}-weight \lambda_{u^{\pm}_{i,1}}.
Now, a direct computation shows that h_i (0).\tau.u^{+}_{i,1} = \lambda_{u^{\pm}_{i,1}}(h_i (0))\tau.u^{+}_{i,1}=\tau.u^{+}_{i,1}. Similarly, we see that h_i (0).\tau.u^{-}_{i,1}= -\tau.u^{-}_{i,1}. It follows from these calculations and Theorem 3 that for every m \in M, we have h_i (0).\left(m \otimes_{\mathbb{C}[S_{\ell}]} u^{\pm}_{i,1} \right) = \pm m \otimes_{\mathbb{C}[S_{\ell}]} u^{\pm}_{i,1}, so m \otimes_{\mathbb{C}[S_{\ell}]} u^{\pm}_{i,1} is in the weight space W_{\lambda_{u^{\pm}_{i,1}}}.
The relations in \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}] show that h_i (k) must take m\otimes_{\mathbb{C}[S_{\ell}]} u^{\pm}_{i,1} to a vector of weight \lambda_{u^{\pm}_{i,1}} since if w has weight \lambda_{u^{\pm}_{i,1}} we must have, for 1\le j \le n, h_j(0)h_i(k).w=h_i(k)h_j(0).w=\lambda_{u^{\pm}_{i,1}}(h_j)h_i(k).w. And since \{ \tau.u^{\pm}_{i,1} \mid \tau \in S_{\ell} \} spans the subspace of V^{\otimes \ell} of its associated weight, then h_i (k) must take m \otimes_{\mathbb{C}[S_{\ell}]} u^{\pm}_{i,1} to a sum of terms of the form m_\tau\otimes \tau.u^{\pm}_{i,1} for some m_\tau (well-defined since W is an \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}]–module). Now, for each m \in M we have:
where m' = \sum_{\tau \in S_{\ell}} m_{\tau}.\tau. Note that this u^{\pm}_{i,1} meets the conditions of Lemma 3.3. Therefore each such m' can be recovered from m' \otimes_{\mathbb{C}[S_{\ell}]} u^{\pm}_{i,1}. Then there exists \alpha_{1,h_i(k)} \in \text{End}_{\mathbb{C}}(M) with the property that, for each m \in M, we can find m' such that m' = \alpha_{1,h_i(k)} (m). Defining \alpha_{j,h_i(k)}(m)=\alpha_{1,h_i(k)}(m.\sigma).\sigma^{-1} we see that
Now assume w has weight \lambda. By relations in \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}], for any w \in W, 1\le i \le n, we have
Thus x_j^{\pm} (k') adds \pm \alpha_j = \pm (\varepsilon_j - \varepsilon_{j+1}) to the weight if j = 1, \ldots , n, or adds \alpha_0 = \mp (\varepsilon_1 - \varepsilon_{n+1}) to the weight if j = 0. In particular, by direct calculation, the weight of x_j^{\pm} (k').u^{\mp}_{j,1} is the same as the weight of u^{\pm}_{j,1}. We conclude that for every m \in M, x_j^{\pm} (k').\left(m \otimes_{\mathbb{C}[S_{\ell}]} u^{\mp}_{j,1} \right) = \sum_{\tau \in S_{\ell}} m_{\tau} \otimes_{\mathbb{C}[S_{\ell}]} \tau.u^{\pm}_{j,1} for some m_{\tau} \in M. An argument similar to the one above proves the existence of the \alpha_{j,x_i^\pm(k)} with the properties in the Lemma statement.
For all z \in \{ h_i (k), x_i^{\pm} (k) \mid i = 0, \ldots , n \text{ and } k \in \mathbb{Z} \}, m \in M (with M as in Lemma 3.4), and \mathbf{v} \in V^{\otimes \ell}, we have
Let z = x_i^{+} (k), i = 2, \ldots , n (in particular, n \geq 2; note that, if n=1, then \ell=1, in which case Lemma 3.5 reduces to Lemma 3.4). Other cases are similar and are omitted.
Let r (resp. s) be the number of tensor factors in \mathbf{v} containing v_i (resp. v_{i+1}). Given \mathbf{v}, write \bar{r}(\mathbf{v}) for the number r in \mathbf{v}, and similarly for \bar{s}, giving s. We will prove the statement of the lemma for all \mathbf{v} \in V^{\otimes \ell} via induction. First induct on r.
Base case: if r=s=0, then the sum \sum_{1 \leq p \leq \ell} \alpha_{p,x_i^{+} (k)}(m) \otimes_{\mathbb{C}[S_{\ell}]} \big(x_i^{+} (k) \big)_p . \mathbf{v} = 0 since x_i^{+} (k).\mathbf{v}=0. Now, x_i^{\pm} (k).(m \otimes_{\mathbb{C}[S_{\ell}]} \mathbf{v}) = \frac{1}{2} \left( h_i(k) x_i^{\pm}(0) - x_i^{\pm}(0) h_i(k) \right).(m \otimes_{\mathbb{C}[S_{\ell}]} \mathbf{v}). By Theorem 3, x_i^{\pm}(0) .(m \otimes_{\mathbb{C}[S_{\ell}]} \mathbf{v}) = m \otimes_{\mathbb{C}[S_{\ell}]} x_i^{\pm}(0).\mathbf{v} = 0, and h_i(k) preserves the weight space of m \otimes_{\mathbb{C}[S_{\ell}]} \mathbf{v}, so x_i^{\pm}(0) applied to h_i (k) (m \otimes_{\mathbb{C}[S_{\ell}]} \mathbf{v}) is 0. Hence both terms are 0, proving the Lemma when r=s=0.
Inductive step: assume that \mathbf{v} is such that \bar{r}(\mathbf{v})=r+1 and \bar{s}(\mathbf{v})=s. Then there exists k_0 \in \{ 1 , \ldots , n \} such that:
v_{k_0} does not appear in \mathbf{v} (since \ell \leq n);
we may assume that k_0 = i-1 (n \geq 2). Otherwise there exists \rho \in W(\mathfrak{sl}_{n+1}), the Weyl group of \mathfrak{sl}_{n+1}, such that \rho(k_0) = i-1 and \rho(i)=i, \rho(i+1)=i+1. Then \bar{r}(\rho(\mathbf{v}))=r+1, \bar{s}(\rho(\mathbf{v}))=s and \mathbf{v} has no copies of v_{i-1}.
If \ell \geq 2, take p such that i_p=i (which is possible if \ell \geq 2, since r+1>0) and define \mathbf{w} such that w_{i_q}=v_{i_q} for q \neq p and w_{i_p}=v_{i-1}. Then \mathbf{v} = x^{-}_{i-1} (0) \mathbf{w} and \bar{r}(\mathbf{w})=r, \bar{s}(\mathbf{w})=s, so we can apply the induction hypothesis to \mathbf{w}. So
which proves the statement of the lemma for all r by induction. (If \ell=1, take \mathbf{v}=v_i and \mathbf{w}=v_{i-1} and perform the same computation above.) A similar argument establishes the statement for all s, completing the proof of the lemma.
For each i,j \in [1,n], p \in [1,\ell], and k \in \mathbb{Z}, we have
where z,z' is x^{\pm} or h.
We first show that \alpha_{p,h_i(k)}=\alpha_{p,h_j(k)}. This is trivial if n=1, so assume n \geq 2. For our first case, assume that \ell \geq 2 and without loss of generality that 1\le i<j\le n. Let
where a_i\in V^{\otimes (\ell-2)} is a pure tensor having indices all distinct and not among \{i,i+1,i+2\}.
We have:
Now act on the right side of (Proof) by x_{i+1}^-(0)x_{i+1}^+(0) to get:
Acting on the left side of (Proof) by x_{i+1}^-(0)x_{i+1}^+(0) gives
and hence \alpha_{p,h_i(k)}(m)=\alpha_{p,h_{i+1}(k)}(m) by Lemma 3.3. Now repeat the same process to get \alpha_{p,h_i(k)}(m)=\alpha_{p,h_j(k)}(m) for 1 \le i<j \le n.
If instead \ell=1, assume 1\le j<i \le n and take \mathbf{v} = v_i. Then compute (Proof) and act by x_{i-1}^-(0)x_{i-1}^+(0) to show by a similar process that \alpha_{p,h_i(k)}(m)=\alpha_{p,h_{i-1}(k)}(m).
Now, we show that \alpha_{p,h_i(k)}=\alpha_{p,x^{\pm}_i(k)}, 1\le i \le n. Let
where b_i\in V^{\otimes (\ell-1)} is a pure tensor having indices all distinct and not among \{i,i+1\}. (If \ell=1, take \mathbf{w} = v_i.) We compute
Again, Lemma 3.3 shows that \alpha_{p,h_i(k)}=\alpha_{p,x^+_i(k)}. A similar argument gives \alpha_{p,h_i(k)}=\alpha_{p,x^-_i(k)}.
We now show by induction the second equality of the lemma statement for the case n \geq 2, k > 0 (the other cases are similar). The base case k=1 is trivial, so assume the statement for k \geq 1. From now on, let \alpha_{p,z_i(k)}=\alpha_{p,k} for 1 \le i \le n. We have
The left hand side above is equal to \alpha_{p,k}\alpha_{p,1}(m) \otimes_{\mathbb{C}[S_{\ell}]} u_{2,p}^+ = \left(\alpha_{p,1} \right)^k \alpha_{p,1}(m) \otimes_{\mathbb{C}[S_{\ell}]} u_{2,p}^+ by the inductive hypothesis, and so the statement follows. By similar arguments, we see that \alpha_{p,z_{0}(k)}=\alpha_{p,1}^k\alpha_{p,z_{0}(0)}, as well as \alpha_{p,-k}=\alpha_{p,-1}^k and \alpha_{p,z_{0}(-k)}=\alpha_{p,-1}^{k}\alpha_{p,z_{0}(0)}, completing the proof.
For each z \in \{ h_i (k), x_j^{\pm} (k) \mid i \in [0,n]; j \in [1, n]; k \in \mathbb{Z} \}, m \in M (with M as in § ?), set m.\mathbf{x}_p^k = \alpha_{p,z}(m). For x_0^{\pm} (k), set m.\mathbf{x}_p^k \mathbf{y}_p^{\pm 1} = \alpha_{p,x_0^{\pm} (k)}(m). Here, \alpha_{p,z} is as in Lemma 3.5. This assignment defines a right \operatorname{Aff}^2(S_{\ell}) \cong \mathbb{C} [(\mathbb{Z}^{\ell} \times \mathbb{Z}^{\ell}) \rtimes S_{\ell}]–module structure on M, extending its \text{Aff}^{1}(S_{\ell}) \cong \mathbb{C} [\mathbb{Z}^{\ell} \rtimes S_{\ell}] structure.
The \text{Aff}^{1}(S_{\ell})–module structure on M is inherited from Theorem 3 (the k=0 case), so we only need to show that the relations from Definition 4 involving \mathbf{x}_j^{\pm 1} in \operatorname{Aff}^2(S_{\ell}) hold. By Lemma 3.3, it is sufficient to show this on m \otimes_{\mathbb{C}[S_{\ell}]} \mathbf{v} for arbitrary m but a particular choice of \mathbf{v} such that \mathbf{v} has \ell distinct components.
All relations except \mathbf{y}_1 \cdots \mathbf{y}_{\ell} \mathbf{x}_1 = \mathbf{x}_1 \mathbf{y}_1 \cdots \mathbf{y}_{\ell} and \mathbf{x}_1 \mathbf{y}_2 = \mathbf{y}_2 \mathbf{x}_1 follow from Lemma 3.6, noting that all h_i (k) are commuting. We show that the remaining two relations hold.
First we show m.\mathbf{y}_2 \mathbf{x}_1 = m.\mathbf{x}_1 \mathbf{y}_2. Note that this relation does not appear if \ell=1, so we will assume \ell \geq 2 (and hence n \geq 2; recall \ell \leq n).
Take \mathbf{v} = v_2 \otimes v_1 \otimes v_3 \otimes v_4 \otimes \cdots \otimes v_{\ell} and \mathbf{v}' = v_2 \otimes v_{n+1} \otimes v_3 \otimes v_4 \otimes \cdots \otimes v_{\ell}. Now
By relations in \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}] we can write the left hand side as
Thus Lemma 3.3 implies that m.\mathbf{y}_2 \mathbf{x}_1 = m.\mathbf{x}_1 \mathbf{y}_2 for all m \in M.
Now we show that m.(\mathbf{y}_1 \cdots \mathbf{y}_{\ell} \mathbf{x}_1) = m.(\mathbf{x}_1 \mathbf{y}_1 \cdots \mathbf{y}_{\ell}). When combined with the previous relation and induction, it is sufficient to demonstrate m.\mathbf{y}_1 \mathbf{x}_1 = m.\mathbf{x}_1 \mathbf{y}_1.
Take \mathbf{v} = v_1 \otimes \cdots \otimes v_{\ell} and \mathbf{v}' = v_{n+1} \otimes v_2 \otimes v_3 \otimes \cdots \otimes v_{\ell}.
Then
By the relations in \mathfrak{sl}_{n+1}[s^{\pm 1}, t^{\pm 1}], this is equal to
Therefore, Lemma 3.3 implies that m.(\mathbf{x}_1 \mathbf{y}_1) = m.(\mathbf{y}_1 \mathbf{x}_1) for all m \in M.
Let \mathfrak g be any perfect Lie algebra (i.e. \mathfrak g = [\mathfrak g,\mathfrak g]). Let L be any Lie algebra over \mathbb{C}. Assume that we have system of Lie algebra homomorphisms \rho_i: L_i(\mathfrak g)\to L such that,
(i) \rho_0:=\rho_i\circ \iota_{i,0}=\rho_j\circ \iota_{j,0},
(ii) for all c \in [1,m] and all subsets \{i_1,\dots, i_c\} of \{1,\dots, m\} we have that
and
(iii)
whenever (j_1,j_2,\dots, j_c) is a permutation of (i_1,i_2,\dots, i_c).
Then there is a unique Lie algebra homomorphism \rho:L^m(\mathfrak g)\to L such that \rho_i=\rho\circ \iota_i.
In this proof, by \rho_i(x) for x \in \mathfrak g we will mean (\rho_i \circ \iota_{i,0})(x). By (i), this means that \rho_i(x)=\rho_j(x). We now prove the statement of the proposition by induction on m. The statement is vacuously true if m=1. For the inductive step, we proceeds as follows.
For all x\in \mathfrak g choose a decomposition x = \sum_{j} [x_{j}', x_{j}''] (a finite sum). We define inductively:
We need to show two facts: (1) \rho is well-defined, i.e. it doesn’t depend on the choice of decomposition, (2) \rho does indeed define a Lie algebra homomorphism. The main step of this argument is to show that \rho, assumed well-defined for m-1, can be expressed in the form:
where
This is clear when m=0,1 so assume that (Proof) holds for m-1, and compute:
where
as desired.
Now, we use (Proof) to prove that \rho is well-defined on L^m(\mathfrak g). In particular, it is independent of the chosen decomposition. Let x=\sum_i [x_{i,1}',x_{i,1}'']=\sum_i[x_{i,2}',x_{i,2}''] be two distinct decompositions and use (Proof) to write
Now, observe that, by our choices
Therefore condition (ii) applies to show:
Hence, the decomposition of x does not affect the definition of \rho.
Now we show that \rho is a homomorphism of Lie algebras. We want to show that
Write \mathbf r = (r_1, \ldots , r_m), \mathbf s = (s_1, \ldots , s_m). We first check the equality in the case when s_j=0 for all j. Here the result holds by definition. Then we assume that this has been proved for s_j=0 for j>p and prove it for s_j=0 with j>p+1. By (Proof) we write:
Then:
using (iii) in the second line to permute the \rho_{i} and powers of t_i. Now, assume \rho is a homomorphism for 1\le p < m. The definition of \rho gives:
An argument similar to the one given above shows that, for all z \in \mathfrak g
Finally, we use the Jacobi Identity and induction to write:
Hence the proposition is proved. The maps and algebras involved are collected in the following commutative diagram.
In case \mathfrak{g} is a finite dimensional semisimple Lie algebra, it can be checked by a similar argument that condition (iii) is irrelevant. We do not show this fact here.
Let A_i for i=1, 2, \ldots , m be a collection of subalgebras of an associative algebra A such that the A_i generate A.
(a) For all A–modules M,N with linear map f : M \to N, the map f is an A-homomorphism if and only if f is an A_i-homomorphism for each i.
(b) Given algebra homomorphisms \phi: A \to B and \phi': A \to B, if \phi|_{A_i}= \phi'|_{A_i} for each i, then \phi = \phi'.
Note that any Lie algebra L can be embedded inside its universal enveloping algebra U(L), and a generating set of L is also a generating set of U(L). In this way Lemma 4.1 also applies to Lie algebras.
Let W be a \mathfrak g–module and assume that there exist Lie algebra homomorphisms \rho_i: L_i(\mathfrak g)\to End W satisfying the conditions of Proposition 8. Then there is a unique L^m(\mathfrak g)–module structure on W whose restriction to L_i via \iota_i is the given L_i(\mathfrak g)–module structure.
The statement is trivial if m=1; we will proceed now by induction. Assume the statement for m=k-1, with 2 \leq k \in \mathbb{Z}, and consider m=k.
Let y \otimes t^{\mathbf p_k} \in L^k(\mathfrak g) where t^{\mathbf p_k} := t_1^{p_1} \cdots t_k^{p_k} be such that \displaystyle y=\sum_{y_1,\dots, y_k\in \mathfrak{g}} [y_k, [y_{k-1}, \cdots[y_2, y_1]\cdots ]]. Let now \displaystyle y'=\sum_{y_1,\dots, y_{k-1}\in \mathfrak{g}} [y_{k-1}, [y_{k-2}, \cdots[y_2, y_1]\cdots ]]. Then for w \in W, define an action of y \otimes t^{\mathbf p_k} on w by (y \otimes t^{\mathbf p_k})w := (y_k \otimes t^{p_k})(y' \otimes t^{\mathbf p_{k-1}})w - (y' \otimes t^{\mathbf p_{k-1}}) (y_k \otimes t^{p_k})w. By the inductive assumption, this action is a well-defined action of L^k(\mathfrak g) on W. That its restriction to L_i via \iota_i is the given L_i(\mathfrak g)–module structure follows from Proposition 8. Uniqueness follows from Lemma 4.1 (b) since the L_i(\mathfrak g), i \in [1,k], generate L^k(\mathfrak g) as a Lie algebra.
Define \operatorname{Aff}^m(S_{\ell}) to be the unital associative algebra over \mathbb{C} with generators \sigma^{\pm 1}_i, i \in [1, \ell-1 ], \mathbf{y}_{1,r}, ..., \mathbf{y}_{m,r}, r \in [1, \ell ] and relations
for k \in [1, \ell-2 ] in the second relation; i, i' \in [1, m]; k,j \in [1, \ell-1 ] otherwise; and p,s \in [1, \ell ].
Let G be any group. For any system of group homomormphisms: \rho_i: \operatorname{Aff}_i(S_{\ell}) \to G satisfying the conditions
(i) \rho_i\circ \nu_0=\rho_j\circ\nu_0, (so denote \rho_0=\rho_i\circ \nu_0: S_{\ell}\to G), and
(ii) \rho_i\ker(ev_i) \rho_j\ker(ev_j)= \rho_j\ker(ev_j) \rho_i\ker(ev_i) for all i, j,
there is a unique group homomorphism \rho: \operatorname{Aff}^m(S_{\ell})\to G such that \rho_i=\rho\circ \nu_i.
We will prove this by induction on m. The statement is obvious when m=1. Assuming the statement for m=k-1 we prove the result for case m=k. Let \sigma \in S_{\ell} and \mathbf r\in \left(\mathbb{Z}^{\ell} \right)^k with \mathbf r = (r_1, \ldots , r_k). Using the inductive assumption, define
The fact that \rho is well-defined, unique, and satisfies the condition \rho_i=\rho\circ \nu_i follow from the definition once we prove that \rho is shown to be a homomorphism of groups. This is done as follows: let \tau\inS_{\ell} and \mathbf q=(q_1,\ldots, q_k). Then
where the third equality follows from the inductive definition of \rho, and the last equality from condition (ii). Now note that, for any j \in [1,k], by condition (i) we have \rho_0(\sigma) \rho_j(q_j,1) = \rho_j(\sigma(q_j),\sigma)=\rho_j(\sigma(q_j),1) \rho_0 (\sigma). Thus (*) is
Let M be an S_{\ell}–module over \mathbb{C}. Assume that there is an \operatorname{Aff}_i(S_{\ell})–module structure on M, denoted by M_i, extending the S_{\ell}–module structure on M. Then there is a unique \operatorname{Aff}^m(S_{\ell})–module structure on M whose restriction to \operatorname{Aff}_i(S_{\ell}) via \nu_i is M_i.
Define an \operatorname{Aff}^m(S_{\ell})–module structure on M by declaring that (n_1, n_2, \ldots , n_m, \sigma) act as (n_1, 1) (n_2, 1) \cdots (n_m, 1) (0, \sigma) with each (n_i,1) denoting the M_i structure and (0 , \sigma) the S_{\ell}–module structure. Since (0, \sigma)(n,1) = (\sigma(n),\sigma) = (\sigma(n),1)(0,\sigma), we have that the product (n_1, \ldots , n_m, \sigma) (n'_1, \ldots , n'_m, \sigma') = (n_1+\sigma(n'_1), \ldots , n_m+\sigma(n'_m), \sigma \sigma'), and thus this action does indeed define an \operatorname{Aff}^m(S_{\ell})–module structure. That its restriction to \operatorname{Aff}_i(S_{\ell}) via \nu_i is M_i is obvious. Uniqueness is a consequence of Lemma 4.1 (b) since the \operatorname{Aff}_i(S_{\ell}) generate \operatorname{Aff}^m(S_{\ell}) as an associative algebra.
Let \ell, m\geq 1 be integers with \mathfrak{g}=\mathfrak{sl}_{n+1} and \ell\leq n. Then
(a) For any \mathbb{C}[\operatorname{Aff}^m(S_{\ell})]–module M there is a unique L^m(\mathfrak g)–module structure on \mathcal F(M)=M\otimes_{\mathbb{C}[S_{\ell}]} V^{\otimes {\ell}} extending the L_i(\mathfrak g)–module structures given by Theorem 3.
(b) \mathcal F: \operatorname{Mod-}\mathbb{C}[\operatorname{Aff}^m(S_{\ell})]\to L^m(\mathfrak{g})\operatorname{-Mod} is a fully faithful functor.
(c) Let L^m(\mathfrak g)\operatorname{-Mod}^{\ell} be the full subcategory of all L^m(\mathfrak g)–modules whose restriction to \mathfrak g via \iota_0^m is locally finite dimensional and all composition factors are the composition factors of V^{\otimes {\ell}}. Then \mathcal F: \operatorname{Mod-}\mathbb{C}[\operatorname{Aff}^m(S_{\ell})]\to L^m(\mathfrak g)\operatorname{-Mod}^{\ell} is a category equivalence.
(a) By Corollary 12 we see that the restricted \operatorname{Aff}_i(S_{\ell})–module structures M_i, 1\le i \le m combine to give a unique \operatorname{Aff}^m(S_\ell)–module structure, which must therefore be M. Let L=\mathfrak{gl}\left(\mathcal{F} (M) \right), and define \rho_i:L_i(\mathfrak g)\to L, 1\le i \le m by \rho_i(x \otimes t_i^{n_i}) (m \otimes \mathbf{v}) = (x \otimes t_i^{n_i}).(m \otimes \mathbf{v}) and check that the \rho_i satisfy the conditions of Proposition 8.
(i) For x \in \mathfrak{g}, \rho_i\circ \iota_{i,0} (x) = \rho_i (x \otimes t_i^0) = \rho_j (x \otimes t_j^0) = \rho_j\circ \iota_{j,0} (x).
(ii) We show this in two steps. First, we show that for any sequence (i_1, \dots, i_c) of elements of \{1, \ldots, m\}, and for any x_1, \ldots, x_c\in \mathfrak g, we have
where \mathbf y_{i,j}^p, 1 \le i \le m, 1\le j \le \ell, p \in \mathbb Z is shorthand for (p\mathbf e_{ij},1)\in \text{Aff}^m(S_\ell) where \mathbf e_{ij} is the m-tuple with the unit vector \mathbf e_j in the ith position and 0’s elsewhere. We use induction on c with the c=1 case following from Theorem 3. We compute:
The last line is 0 unless j=k, in which case it becomes:
It follows that
if
(iii) As in the previous step, we see that
If (j_1,\dots, j_c) is a permutation of (i_1,\dots i_c) then
which is the same element of M\otimes V^{\otimes \ell} since the \mathbf y_{i,j} commute.
Therefore, the \rho_i thus defined satisfy the conditions of Proposition 8. The existence of the L^m(\mathfrak g)–module structure follows from Proposition 8, and uniqueness follows from Lemma 4.1.
(b) As the functor is fully faithful in the m=1 case, this statement follows from Theorem 3, Proposition 8, and Lemma 4.1.
(c) We have the restrictions \mathcal{F}(M) \to \left( \mathcal{F}(M)|_{L_i (\mathfrak{g})} \right) \big|_{\mathfrak{g}} via \iota_0^m which each satisfy the stated conditions. Then the statement follows from Theorem 3, Proposition 8, and Lemma 4.1.
The text on this page is from arXiv:2407.08004v1 by Vyjayanthi Chari, Lauren Grimley, Zongzhu Lin, Chad R. Mangum, Christine Uhl and Evan Wilson, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2407/2407.08004.jsonl.