The geometric \(R\)-matrix for affine crystals of type \(A\) (Q2237391)

Quantum affine algebras admit a class of finite-dimensional, non-highest weight representations called Kirillov-Reshetikhin (KR) modules. The crystal bases of these representations, which we call KR crystals, have received a lot of attention for several reasons. Kang et al. showed that the crystal bases of highest-weight modules for quantum affine algebras can be built out of infinite tensor products of KR crystals, and they used this construction to compute the one-point functions of certain solvable lattice models coming from statistical mechanics [\textit{S.-J. Kang} et al., Int. J. Mod. Phys. A 7, 449--484 (1992; Zbl 0925.17005)], [\textit{D. Takahashi} and \textit{J. Satsuma}, ``A soliton cellular automaton'', J. Phys. Soc. Japan 59, No. 10, 3514--3519 (1990); \textit{G. Hatayama} et al., J. Math. Phys. 42, No. 1, 274--308 (2001; Zbl 1032.17021)]. In the case of KR crystals, however, there is a unique crystal isomorphism \[ \widetilde{R} : B_1 \otimes B_2 \rightarrow B_2 \otimes B_1. \] This isomorphism is the combinatorial \(R\)-matrix, and it plays an essential role in the sequel. For example, the states of the box-ball system can be represented as elements of a tensor product of KR crystals, and the time evolution is given by applying a sequence of combinatorial \(R\)-matrices. In affine type \(A\), Kirillov-Reshetikhin modules correspond to partitions of rectangular shape \((L^k)\), and their crystal bases, which we denote by \(B^{k,L}\), are modeled by semistandard Young tableaux of shape \((L^k)\). If one ignores the affine crystal operators \(\widetilde{e}_0, \widetilde{f}_0\), then \(B^{k,L}\) is the crystal associated to the irreducible \(\mathfrak{sl}_n\)-module of highest weight \((L^k)\). Shimozono showed that the affine crystal operators are obtained by conjugating the crystal operators \(\widetilde{e}_1, \widetilde{f}_1\) by Schützenberger's promotion map. He also gave a combinatorial description of the action of the combinatorial \(R\)-matrix on pairs of rectangular tableaux. Let \(*\) denote the associative product on the set of semistandard Young tableaux introduced by Lascoux and Schützenberger. If \(T \in B^{k_1, L_1}\) and \(U \in B^{k_2, L_2}\), then there are unique tableaux \(U' \in B^{k_2,L_2}\) and \(T' \in B^{k_1,L_1}\) such that \(T*U = U'*T'\), and the combinatorial \(R\)-matrix acts as the map \(\widetilde{R} : T \otimes U \mapsto U' \otimes T'\). Suppose \(T\) and \(U\) are one-row tableaux, with entries at most \(n\), and suppose \(\widetilde{R}(T \otimes U) = U' \otimes T'\). Let \(a_j, b_j\) be the numbers of \(j\)'s in \(T\) and \(U\), respectively. Define \[ b'_j = b_j + \widetilde{\kappa}_{j+1} - \widetilde{\kappa}_j, \quad\quad a'_j = a_j + \widetilde{\kappa}_j - \widetilde{\kappa}_{j+1}, \] where \[ \widetilde{\kappa}_j = \min_{0 \leq r \leq n-1} (b_j + b_{j+1} + \cdots + b_{j+r-1} + a_{j+r+1} + a_{j+r+2} + \cdots + a_{j+n-1}),\tag{1} \] and all subscripts are interpreted modulo \(n\). Then \(b'_j, a'_j\) are the numbers of \(j\)'s in \(U'\) and \(T'\), respectively. A geometric (or rational) lift of a piecewise-linear function \(\widetilde{h}\) is any positive rational function \(h\) which tropicalizes to \(\widetilde{h}\). Rational functions are often easier to work with than piecewise-linear functions, since one may bring to bear algebraic and geometric techniques. Furthermore, identities proved in the lifted setting can be ``pushed down,'' via tropicalization, to results about the piecewise-linear functions and the corresponding combinatorial maps. For example, the formula for \(\widetilde{R}\) in Proposition 1 turns out to be the tropicalization of a rational map which solves a certain matrix equation. Given \(x = (x_1, \ldots, x_n) \in (\mathbb{C}^\times)^n\), define \[ g(x) = \left( \begin{array}{ccccccc} x_1 & &&&& \lambda \\ 1 & x_2 & \\ & 1 & x_3 & \\ & & & \ddots \\ & & & & x_{n-1} \\ & & & & 1 & x_n \end{array} \right). \] Here \(\lambda\) is an indeterminate, and we view \(g(x)\) as an element of the loop group \(\mathrm{GL}_n(\mathbb C(\lambda))\). Proposition 2. If \(x,y \in (\mathbb{C}^\times)^n\) are sufficiently generic, then the matrix equation \[ g(x)g(y) = g(y')g(x') \] has two solutions: the trivial solution \(y'_j = x_j, x'_j = y_j\), and the solution \[ y'_j = y_j\dfrac{\kappa_{j+1}}{\kappa_j}, \quad x'_j = x_j\dfrac{\kappa_j}{\kappa_{j+1}}, \quad \text{ where } \quad\quad \kappa_j = \sum_{r=0}^{n-1} y_j \cdots y_{j+r-1} x_{j+r+1} \cdots x_{j+n-1},\tag{2} \] and subscripts are interpreted modulo \(n\). The solution given by (2) is the unique solution to (2) which satisfies the additional constraint \[ \prod x_j = \prod x_j' \quad\quad \text{ and } \quad\quad \prod y_j = \prod y_j'.\tag{3} \] Note that the piecewise-linear map \(\widetilde{R}\) in Proposition (2) is the tropicalization of the rational map \(R : (x,y) \mapsto (y',x')\), where \(y',x'\) are defined by (2) (note also that (3) tropicalizes to the condition \(\sum a_j = \sum a'_j, \sum b_j = \sum b_j'\), which says that the tableaux \(T\) and \(T'\) (resp., \(U\) and \(U'\)) have the same length). Thus, the map \(R\) is a geometric lift of the combinatorial \(R\)-matrix on pairs of one-row tableaux. One of the crowning achievements of the geometric lifting program is Berenstein and Kazhdan's theory of geometric crystals, which provides a framework for lifting the entire combinatorial structure of crystal bases [\textit{A. Berenstein} and \textit{D. Kazhdan}, in: GAFA 2000. Visions in mathematics---Towards 2000. Proceedings of a meeting, Tel Aviv, Israel, August 25--September 3, 1999. Part I. Basel: Birkhäuser. 188--236 (2000; Zbl 1044.17006); Contemp. Math. 433, 13--88 (2007; Zbl 1154.14035); MSJ Mem. 17, 1--9 (2007; Zbl 1146.17306)]. Roughly speaking, a geometric crystal is a complex algebraic variety \(X\), together with rational actions \(e_i : Cx \times X \rightarrow X\), which are called geometric crystal operators. The geometric crystal operators are required to satisfy rational lifts of the piecewise-linear relations satisfied by (combinatorial) crystal operators. In many cases, the geometric crystal operators are positive, and they tropicalize to piecewise-linear formulas for the combinatorial crystal operators \(\widetilde{e}_i\) on a corresponding combinatorial crystal \(B_X\); when this happens, we say that \(X\) tropicalizes to \(B_X\). For each reductive group \(G\), \textit{A. Berenstein} and \textit{D. Kazhdan} [Contemp. Math. 433, 13--88 (2007; Zbl 1154.14035)] constructed a geometric crystal on the flag variety of \(G\) which lifts the crystals associated to all the irreducible representations of \(G^\vee\), the Langlands dual group. These geometric crystals provide a new method for constructing and studying crystals, [\textit{T. Lam} and \textit{N. Templier}, ``The mirror conjecture for minuscule flag varieties'', Preprint, \url{arXiv 1705.00758}; \textit{R. Chhaibi}, ``Littelmann path model for geometric crystals, Whittaker functions on Lie groups and Brownian motion'', Preprint, \url{arXiv 1302.0902}; \textit{A. Braverman} et al., in: GAFA 2000. Visions in mathematics---Towards 2000. Proceedings of a meeting, Tel Aviv, Israel, August 25--September 3, 1999. Part I. Basel: Birkhäuser. 237--278 (2000; Zbl 1004.11026); \textit{A. Berenstein} and \textit{D. Kazhdan}, MSJ Mem. 17, 1--9 (2007; Zbl 1146.17306)]. In [\textit{G. Frieden}, ``Affine type \(A\) geometric crystal on the Grassmannian'', Preprint, \url{arXiv:1706.02844}], we constructed a geometric crystal on the variety \(\mathbb X{k} := \mathbb Y{k}{n}\) which tropicalizes to the affine crystal structure on rectangular tableaux with \(n-k\) rows. In this sequel, we define and study the \textit{geometric \(R\)-matrix}, a birational map \(R : \mathbb X{k_1} \times \mathbb X{k_2} \rightarrow \mathbb X{k_2} \times \mathbb X{k_1}\) which tropicalizes to the combinatorial \(R\)-matrix on pairs of rectangular tableaux. The combinatorial \(R\)-matrix Given two type \(A_{n-1}^{(1)}\) crystals \(A,B\), their tensor product \(A \otimes B\) is defined as follows. The underlying set is the Cartesian product \(A \times B\), whose elements we denote by \(a \otimes b\). The crystal structure is defined by \begin{itemize} \item \(\gamma(a \otimes b) = \gamma(a) + \gamma(b)\); \item \(\epsilon_i(a \otimes b) = \epsilon_i(b) + \max(0, \epsilon_i(a) - \phi_i(b))\); \item \(\phi_i(a \otimes b) = \phi_i(a) + \max(0, \phi_i(b) - \epsilon_i(a))\); \item \(e_i(a \otimes b) = \begin{cases} e_i(a) \otimes b & \text{ if } \epsilon_i(a) > \phi_i(b) \\ a \otimes e_i(b) & \text{ if } \epsilon_i(a) \leq \phi_i(b). \end{cases} \) \end{itemize} In the definition of \(e_i\), we take \(0 \otimes b = a \otimes 0 = 0\). The tensor product of crystals is associative, but not commutative. In the case of the Kirillov-Reshetikhin crystals \(B^{k,L}\), however, there is a unique crystal isomorphism \(\widetilde{R} : B^{k_1, L_1} \otimes B^{k_2, L_2} \rightarrow B^{k_2,L_2} \otimes B^{k_1,L_1}\), called the \textit{combinatorial \(R\)-matrix}. The existence and uniqueness of this isomorphism is proved using quantum groups (see [\textit{M. Shimozono}, J. Algebr. Comb. 15, No. 2, 151--187 (2002; Zbl 1106.17305), Thm 3.19]). Lascoux and Schützenberger introduced an associative \textit{tableau product} on the set of semistandard tableaux. Given two tableaux \(T\) and \(U\), the product \(T * U\) may be defined as the rectification of the skew-tableau obtained by placing \(U\) to the northeast of \(T\). Proposition 3. Suppose \((T,U) \in B^{k_1,L_1} \otimes B^{k_2,L_2}\). \begin{itemize} \item[1.] There is a unique pair \((U',T') \in B^{k_2,L_2} \times B^{k_1,L_1}\) such that \(T*U = U'*T'\). \item[2.] The combinatorial \(R\)-matrix is given by \(\widetilde{R}(T,U) = (U',T')\). \end{itemize} Proposition 4. Let \(A,B,C\) be three Kirillov-Reshetikhin crystals of type \(A_{n-1}^{(1)}\). \begin{itemize} \item[1.] The map \(\widetilde{R}^2 : A \otimes B \rightarrow A \otimes B\) is the identity. \item[2.] The combinatorial \(R\)-matrix satisfies the Yang-Baxter relation. That is, if \(\widetilde{R}_1 : A \otimes B \otimes C \rightarrow B \otimes A \otimes C\) is the map which applies \(\widetilde{R}\) to the first two factors and does nothing to the third factor, and \(\widetilde{R}_2 : A \otimes B \otimes C \rightarrow A \otimes C \otimes B\) is the map which applies \(\widetilde{R}\) to the last two factors and does nothing to the first factor, then \[ \widetilde{R}_1 \circ \widetilde{R}_2 \circ \widetilde{R}_1 = \widetilde{R}_2 \circ \widetilde{R}_1 \circ \widetilde{R}_2 \] as maps from \(A \otimes B \otimes C \rightarrow C \otimes B \otimes A\). \end{itemize} Define the \textit{cyclic shift map} \(PR :X{k} \rightarrow X{k}\) by \(PR(N|t) = N'|t\), where \(N'\) is obtained from \(N\) by shifting the rows down by 1 (mod \(n\)), and multiplying the new first row by \((-1)^{k-1} t\). We write \(PR_t\) to denote the map \(N \mapsto N'\). It is easy to see that \(PR\) is well-defined (that is, the definition does not depend on the choice of matrix representative for the subspace \(N\)), and that it has order \(n\). Note that the Plücker coordinates of \(N' = PR_t(N)\) are given by \[ P_J(N') = \begin{cases} P_{J - 1}(N) & \text{ if } 1 \not \in J \\ t \cdot P_{J - 1}(N) & \text{ if } 1 \in J \end{cases}\tag{4} \] where \(J - 1\) is the subset obtained from \(J\) by subtracting 1 from each element (mod \(n\)). Extend \(PR\) to a map \(X{k_1, \ldots, k_d} \rightarrow X{k_1, \ldots, k_d}\) by \[ PR(x_1, \ldots, x_d) = (PR(x_1), \ldots, PR(x_d)). \] Define the \textit{shift map} \(sh\) on an \(n\)-periodic matrix \(X\) by \[ sh(X)_{ij} = X_{i-1,j-1}. \] This map is easily seen to be an automorphism of order \(n\). It is the ``unipotent analogue'' of \(PR\) by the following result. We have the identity \(g \circ PR =sh \circ \, g\) of rational maps \(X{k_1, \ldots, k_d} \rightarrow B^-\). For \(z \in C\), define \(\pi^k_z : M_n(\mathbb C[\lambda, \lambda^{-1}]) \rightarrow X{k}\) by \[ \pi^k_z(A) = N|z\tag{5} \] where \(N\) is the subspace spanned by the first \(k\) columns of the \(n \times n\) matrix \(A_z = A|_{\lambda = (-1)^{k-1} z}\). This is a rational map: it is undefined if the first \(k\) columns of \(A_z\) do not have full rank. Define the \textit{flip map} \(fl\) on an \(n \times n\) matrix \(A\) by \[ fl(A)_{ij} = A_{n-j+1,n-i+1}. \] In words, \(fl\) reflects the matrix over the anti-diagonal. It is easy to see that \(fl\) is an anti-automorphism. Define the \textit{geometric Schützenberger involution} \(S : X{k} \rightarrow X{k}\) by \[ S(N|t) = \pi^k_t \circ fl \circ \, g(N|t). \] Continuing the notation used for \(PR\), we write \(S_t\) to denote the map \(N \mapsto N'\), where \(N'|t = S(N|t)\). The map \(S\) is an involution by [\textit{G. Frieden}, J. Comb. Theory, Ser. A 167, 499--563 (2019; Zbl 1467.17016), Cor. 7.4(1)]. The Plücker coordinates of \(S_t(N)\) will appear frequently enough that one introduces the following: \[ Q^J_t(N) := P_{w_0(J)}(S_t(N)),\tag{6} \] where \(w_0(J)\) is the subset obtained by replacing each \(j \in J\) with \(n-j+1\). The Plücker coordinates \(P_J(N)\) (resp., \(Q^J_t(N)\)) are the maximal minors of the first \(k\) columns (resp., last \(k\) rows) of \(g(N|t)\). Since the bottom left \(k \times k\) submatrix of \(g(N|t)\) is upper uni-triangular, we have \[ \dfrac{P_J(N)}{P_{[n-k+1,n]}(N)} = \Delta_{J,[k]}(g(N|t)) \quad \text{ and } \quad \dfrac{Q^J_t(N)}{Q^{[k]}_t(N)} = \Delta_{[n-k+1,n],J}(g(N|t)).\tag{7} \] Extend \(S\) to a map \(X{k_1, \ldots, k_d} \rightarrow X{k_d, \ldots, k_1}\) by \[ S(x_1, \ldots, x_d) = (S(x_d), \ldots, S(x_1)).\tag{8} \] Note that the order of the factors is reversed. We have the identity \(g \circ S = fl \circ \, g\) of rational maps \(X{k_1, \ldots, k_d} \rightarrow B^-\). Fix \(\ell, k \in [n-1]\). Consider the unipotent crystals \((X{\ell}, g)\) and \((X{k},g)\) and their product is the unipotent crystal \((X{\ell} \times X{k}, g)\), where \(g(u,v) = g(u)g(v)\) for \((u,v) \in X{\ell} \times X{k}\). Recall the geometric Schützenberger involution \(S\) and the ``evaluation-projection'' \(\pi^k_z\). Define a rational map \(\Psi_{k,\ell} : X{\ell} \times X{k} \rightarrow X{k}\) by \[ \Psi_{k,\ell}(M|s, N|t) = \pi^k_t \circ \, g(M|s, N|t). \] The \textit{geometric \(R\)-matrix} is the rational map \(R : X{\ell} \times X{k} \rightarrow X{k} \times X{\ell}\) defined by \[ R = (\Psi_{k,\ell}, \; S \circ \Psi_{\ell,k} \circ S). \] More explicitly, if \(R(M|s, N|t) = (N'|t, M'|s)\), then we have \[ N' = g(M|s)|_{\lambda = (-1)^{k-1}t} \cdot N \quad\quad \text{ and } \quad\quad S_s(M') = fl (g(N|t))|_{\lambda = (-1)^{\ell-1} s} \cdot S_s(M). \tag{9} \] The two crucial results about \(R\) are the following. The geometric \(R\)-matrix is positive. (In the sense discussed in the text) We have the identity \(g \circ R = g\) of rational maps \(X{\ell} \times X{k} \rightarrow B^-\). That is, if \(R(u,v) = (v',u')\) and \(g(v'), g(u')\) are defined, then \[ g(u)g(v) = g(v')g(u'). \] We have the identity \(R^2 =Id\) of rational maps from \(X{\ell} \times X{k}\) to itself. Suppose \({k} = (k_1, \ldots, k_d) \in [n-1]^d\). For \(i = 1, \ldots, d-1\), let \(\sigma_i({k}) = (k_1, \ldots, k_{i+1},k_i, \ldots, k_d)\), and let \[ R_i :X{k} \rightarrow X{\sigma_i({k})} \] be the map which acts as the geometric \(R\)-matrix on factors \(i\) and \(i+1\), and as the identity on the other factors. Say that a point \(N|t \in X{k}\) is \textit{positive} if \(t > 0\), and \(P_J(N) > 0\) for all \(J\). Let \(\mathbb{U}_{{k}}\) (and \(\mathbb{U}_{k_1, \ldots, k_d}\)) denote the subset of \(X{{k}}\) consisting of \((N_1|t_1, \ldots, N_d|t_d)\) such that each \(N_i|t_i\) is positive, and the \(t_i\) are distinct. Note that \(g\) is defined on \(\mathbb{U}_{{k}}\), and since the geometric \(R\)-matrix is positive and involutive, each \(R_i\) is a bijection from \(\mathbb{U}_{{k}}\) to \(\mathbb{U}_{\sigma_i({k})}\). \begin{itemize} \item[1.] \(R : X{k_1} \times X{k_2} \rightarrow X{k_2} \times X{k_1}\) is an isomorphism of geometric crystals. \item[2.] \(R : X{k_1} \times X{k_2} \rightarrow X{k_2} \times X{k_1}\) commutes with the symmetries \(PR, S,\) and \(D\). \item[3.] \(R\) satisfies the Yang-Baxter relation. That is, we have the equality \[ R_1R_2R_1 = R_2R_1R_2\tag{10} \] of rational maps \(X{k_1,k_2,k_3} \rightarrow X{k_3,k_2,k_1}\). \end{itemize} The map \(\Theta R : T{k_1} \times T{k_2} \rightarrow T{k_2} \times T{k_1}\) is positive, so we may define \[ \widehat{R} = Trop(\Theta R) : T{k_1} \times T{k_2} \rightarrow T{k_2} \times T{k_1}. \] If \(a\) is a \(k_1\)-rectangle and \(b\) is a \(k_2\)-rectangle, then \(\widehat{R}(a \otimes b) = \widetilde{R}(a \otimes b)\), where \(\widetilde{R}\) is the combinatorial \(R\)-matrix.