\documentclass[11pt,reqno]{amsart} \usepackage[a4paper,margin=1.1in]{geometry} \usepackage{amsmath,amssymb,mathtools,mathrsfs,booktabs} \usepackage{microtype} \usepackage{xcolor} \usepackage[hidelinks]{hyperref} \hypersetup{colorlinks=true,linkcolor={red!50!black},citecolor={blue!50!black},urlcolor={blue!75!black}, pdftitle={Cluster--crystal realization of KR Cartan algebras}, pdfauthor={Jiarui Fei}} \setlength{\emergencystretch}{2em} \numberwithin{equation}{section} \DeclareMathOperator{\wt}{wt} \DeclareMathOperator{\Cox}{Cox} \DeclareMathOperator{\Cl}{Cl} \DeclareMathOperator{\Spec}{Spec} \DeclareMathOperator{\Frac}{Frac} \DeclareMathOperator{\supp}{supp} \DeclareMathOperator{\ord}{ord} \DeclareMathOperator{\diag}{diag} \DeclareMathOperator{\KR}{KR} \renewcommand{\thepart}{\Roman{part}} \newcommand{\C}{\mathbb C} \newcommand{\Z}{\mathbb Z} \newcommand{\Up}{\mathcal U_{\mathrm{pol}}} \newcommand{\Lp}{\mathcal L_{\mathrm{pol}}} \newcommand{\g}{\mathfrak g} \newcommand{\ga}{\mathfrak a} \newcommand{\h}{\mathfrak h} \newcommand{\tc}{\mathfrak C} \newcommand{\cL}{\mathcal L} \DeclareMathOperator{\Proj}{Proj} \DeclareMathOperator{\Pic}{Pic} \DeclareMathOperator{\Ad}{Ad} \DeclareMathOperator{\divg}{div} \DeclareMathOperator{\Mat}{Mat} \DeclareMathOperator{\Res}{Res} \DeclareMathOperator{\Der}{Der} \newcommand{\cA}{\mathcal A} \newcommand{\cB}{\mathcal B} \newcommand{\cD}{\mathcal D} \newcommand{\cJ}{\mathcal J} \newcommand{\OO}{\mathcal O} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \title[Cluster--crystal realization of KR Cartan algebras]{Cluster--Crystal Realization of KR Cartan Algebras} \author{Jiarui Fei} \address{School of Mathematical Sciences, Shanghai Jiao Tong University, Shanghai, China} \email{jiarui@sjtu.edu.cn} \subjclass[2020]{Primary 13F60, 17B67; Secondary 14M17, 14L30} \keywords{Schubert Cox ring, upper cluster algebra, Cartan algebra, Kirillov--Reshetikhin module, theta basis, crystal, polyhedral model} \thanks{The author was supported in part by the National Natural Science Foundation of China (Nos.~12571038).} \date{} \begin{document} \raggedbottom \begin{abstract} We realize the Cartan algebras of restricted Kirillov--Reshetikhin modules as Cox rings of affine Schubert varieties, with Cartan degree given by the primitive divisor class. This yields upper cluster structures with non-inverted frozen variables for untwisted and diagram-twisted current algebras. In untwisted type, translation of the loop parameter gives a second cluster structure whose frozen variables are indexed by the affine Dynkin nodes. We construct charts adapted to the finite Chevalley operators and both shifted affine simple-root operators. For every untwisted finite simple type, every fundamental node, and every Cartan degree, the translated theta basis carries a finite regular affine crystal. The construction uses a uniform reddening theorem and perfectness of the current operators on theta functions. A degree--level identity obtained from tropical Donaldson--Thomas duality determines the minimum level in divisible degrees and gives a finite minimization formula in all degrees. The crystals are simple, their tensor products are connected, and the divisible-degree crystals have a common limit. The affine theta crystals have the finite characters of the corresponding restricted KR modules; they are identified with quantum KR crystals in every classical type and at the exceptional minuscule and adjoint nodes covered by the available recognition theorems. \end{abstract} \maketitle \enlargethispage{2pt} \setcounter{tocdepth}{1} \tableofcontents \clearpage \section*{Introduction} Fix a fundamental node $r$ of a finite-dimensional simple Lie algebra $\g$. For $s\ge0$, let $K_{r,s}$ be the \emph{restricted Kirillov--Reshetikhin module} of highest $\g$-weight $s\omega_r$. The cyclic tensor maps $K_{r,s+t}\to K_{r,s}\otimes K_{r,t}$ define the graded dual \emph{KR Cartan algebra} \[ R_r=\bigoplus_{s\ge0}K_{r,s}^*. \] We construct cluster coordinates that respect Cartan multiplication and express the current operators as derivations in adapted charts. The KR--Demazure correspondence \cite{CM,FL,FK,KV} realizes suitable Veronese subalgebras of $R_r$ as section rings on affine Schubert varieties $Y_r$. To recover the full algebra, we use a finite jet group acting on every $K_{r,s}$. A common subgroup stabilizes the cyclic lines through powers of one character. Matrix coefficients identify $R_r$ with the invariants under the character kernel. We obtain \[ R_r\simeq\Cox(Y_r),\qquad \Cl(Y_r)=\Z[D_r], \] with Cartan degree equal to the primitive divisor class grading (Theorem~\ref{thm:kr-primitive-cox}). This applies to untwisted currents and to the twisted currents defined by diagram automorphisms. For $A_{2n}^{(2)}$, the Demazure polarization has central level two and Veronese index two or four; the Cartan grading remains primitive. The Schubert Cox theorem \cite{FeiSchubert} supplies an upper cluster structure on $R_r$. If the Schubert element $w$ has length $d$ and support $S$, its reduced-word seed has $d-|S|$ mutable variables and $|S|+1$ frozen variables. The latter include the canonical section $\tau$ of the Schubert divisor. The finite Chevalley generators and every nonzero positive affine simple generator have \emph{adapted charts} in this mutation class: \[ D=c_D\prod_{x\ne b(D)}x^{m_x(D)}\partial_{b(D)},\qquad c_D\in\C^\times,\quad m_x(D)\in\Z_{\ge0}, \] where $b(D)$ is frozen. The exchange matrix determines the mutable exponents. We recall the required Schubert results in Section~\ref{sec:schubert-foundation}. For untwisted currents, we construct a second cluster structure. Let $n$ be the finite rank and $L=t_r=2/(\alpha_r,\alpha_r)$, with long roots of squared length two. Translation of the loop parameter by $a\ne0$ gives a homogeneous seed $\Sigma_a$ with $d-n$ mutable variables and frozen coordinates $b_0(a),\ldots,b_n(a)$. If $k_i$ are the affine comarks, Theorem~\ref{thm:untwisted-translated-cluster} gives \[ R_r=\Up(\Sigma_a),\qquad \deg b_i(a)=Lk_i\quad(0\le i\le n). \] Outside the evaluation cases, the Schubert seed has full affine support and $n+2$ frozen variables. The translated seed has $n+1$, since $\tau$ becomes mutable. We recover its exchange from a Poisson bracket for which $\tau=0$ is not a Poisson divisor. The residue of the translation field is a simple-root operator on the inverse Schubert boundary cell. Its root strings compute the pole orders of the translated minors and hence the frozen degrees required by Proposition~\ref{prop:volume-criterion}. The translated mutation class has charts adapted to every finite Chevalley generator and to $F_0(a)=e_\theta\otimes(u+a)^{-1}$, where the inverse is taken in the common finite jet algebra. The same holds for $E_0(a)=f_\theta\otimes(u+a)$. At the exceptional nonneutral nodes this follows by transporting a current chart through a common spectral cluster (Theorem~\ref{thm:untwisted-uniform-currents}). For the distinguished lowering operator $F_r$, we reverse the word presentation on the Schubert boundary cell, apply the simple-root mutations, and then mutate $\tau$. In the evaluation cases the two cluster structures coincide. In type $A$, the Cartan algebras are Grassmannian coordinate rings, whose cluster structures were constructed by Scott~\cite{Scott}. Section~\ref{sec:examples} gives explicit realizations for the first node of type~$C_n$ and for the even twisted current algebra of type~$A_2^{(2)}$. \subsection{Theta bases and crystals} The Schubert seeds admit maximal green sequences. Their regular theta functions therefore form a homogeneous basis of $R_r$, and the adapted current charts make this basis perfect for all finite Chevalley operators \cite{FeiTheta}. Theorem~\ref{thm:origin-theta-crystals} identifies its homogeneous crystals with the finite-type crystals of the restricted KR modules, uniformly in the untwisted and twisted cases. Every translated seed admits a reddening sequence (Theorem~\ref{thm:translated-reddening}). The accompanying supplement~\cite{FeiReddening} proves the comparison with the spectral seed and the uniform two-strip construction at the double bond. A comparison of symmetric scattering diagrams and explicit Baxter and Toda recurrences give the boundary estimates. The adapted charts make the theta basis perfect for every lowering current, including $F_0(a)$. The rank-one relations give perfectness for the raising currents and identify the leading maps as partial inverses. We obtain finite regular affine crystals with the finite characters of the restricted KR modules (Theorem~\ref{thm:translated-theta-crystals}). We relate the Cartan grading to their minimum levels. Write a regular theta function as $\vartheta_g=b^{\mathbf q}\chi_p$, where $b^{\mathbf q}$ is a frozen monomial and $\chi_p$ has mutable $g$-vector $p$ and vanishing order zero along every frozen divisor. Set $d(p)=\deg\chi_p$, and let $\mathsf S$ be the mutable tropical Donaldson--Thomas map specified in Section~\ref{sec:theta-levels}. Theorem~\ref{thm:theta-degree-level} gives the identity \[ L\ell(g)=\deg\vartheta_g+d(\mathsf S(p)),\qquad \ell(g)=\sum_i k_i\varepsilon_i(g). \] It follows that $L\ell(g)\ge\deg\vartheta_g$, with equality precisely for frozen monomials. In degree $L\ell$, these are exactly the minimal elements, and both string maps identify them with the dominant classical affine weights of level $\ell$. In the other degrees we obtain a finite minimization formula, together with a sharp degree bound for simultaneous kernels of the current operators. Cartan degree and affine level are thus distinct numerical invariants. An important feature of this construction is that neither the affine crystal operators nor the degree--level formula requires an identification with a quantum KR crystal. In particular, the same construction applies at exceptional nodes for which no such identification is established here. Thus the resulting crystals give uniform affine combinatorics for the restricted KR modules in all untwisted types and in every degree. The essential simple socle of the dual current module implies simplicity of the affine crystals and connectedness of their tensor products. Translation by frozen exponent vectors gives a common limit for the degrees $L\ell$ on the lattice of Cartan degree zero. In classical types, a uniqueness theorem identifies these crystals with $B^{r,s}$; they are perfect exactly when $L\mid s$. We also prove the quantum identification at the exceptional minuscule and adjoint nodes (Corollary~\ref{cor:theta-exceptional-quantum}). \subsection{Relation to earlier work} The restricted KR presentations are due to Chari and Moura~\cite{CM}. We use the KR--Demazure correspondence of Fourier--Littelmann in the untwisted case~\cite{FL}, and Fourier--Kus and Kus--Venkatesh in the twisted cases~\cite{FK,KV}. These module identifications respect the cyclic tensor maps and hence the Cartan multiplication. The reduced-word seeds come from Goodearl--Yakimov's integral quantum cluster structures~\cite{GYIntegral}. Their Cox-ring lifts are related to the constructions of Gei\ss--Leclerc--Schr\"oer and Francone \cite{GLSFlags,FranconeLifting,Francone}. We use the Cox-ring equality, root-operator charts, and the criterion based on canonical degree of~\cite{FeiSchubert} to identify the primitive KR grading and construct the translated seed. The Poisson calculations use Polishchuk's canonical Poisson-module connection~\cite{PolishchukPoisson} and the rational and trigonometric loop-group structures~\cite{KWWY,FKR,AbedinBurban}. We use the theta functions of Gross--Hacking--Keel--Kontsevich~\cite{GHKK}. For the crystal construction, theta perfectness, frozen valuations, and tropical Donaldson--Thomas duality come from~\cite{FeiTheta}. The quantum alcove and quantum Lakshmibai--Seshadri path models give uniform realizations of single-column quantum KR crystals in all untwisted types~\cite{LNSSS}. General higher-level models and their relation to Demazure crystals have been studied by Lenart--Scrimshaw~\cite{LenartScrimshaw}; recent work gives a uniform PBW model at special nodes~\cite{JangScrimshaw}. Our construction is different: its theta functions and crystal operators arise from the KR Cartan algebra, and the comparison with quantum KR crystals is proved in the cases stated above. Berenstein--Kazhdan's comparison theorem~\cite{BK} gives finite-type regularity. We use simple crystals and limits of coherent families as in Akasaka--Kashiwara~\cite{AK} and Kang--Kashiwara~\cite{KKHead}, and the classical recognition and perfectness theorems of~\cite{ST,FOSPerfect}. \subsection{Conventions} We work over $\C$ and use Bourbaki numbering for finite Dynkin nodes. We write $\Up(\Sigma)$ for the upper cluster algebra with non-inverted frozen variables. Our exchange matrices have mutable indices on their rows and all seed indices on their columns, so mutation at $k$ is given by \[ x_kx_k'=\prod_jx_j^{[B_{kj}]_+}+\prod_jx_j^{[-B_{kj}]_+},\qquad [c]_+=\max(c,0). \] \part{Upper cluster structures} \section{Schubert Cox rings}\label{sec:schubert-foundation} We recall the Schubert Cox results of \cite{FeiSchubert} in the form needed here. Let $Y=X_w^{P_J}$ be a positive-dimensional Schubert variety in a partial flag variety of a symmetrizable Kac--Moody group, where $w$ is the minimal representative of its coset in $W/W_J$. Write $C$ for its open cell, $d=\ell(w)$, and $S=\supp(w)$. If $D_1,\ldots,D_q$ are the prime divisors in $Y\setminus C$, their classes form a basis of $\Cl(Y)$. We use these divisors to define the \emph{Cox ring} \[ \Cox(Y)=\bigoplus_{\mathbf a\in\Z^q}H^0\!\left(Y,\OO_Y\!\left(\sum_{j=1}^q a_jD_j\right)\right)\boldsymbol\tau^{\mathbf a}\subset\C[C][\tau_1^{\pm1},\ldots,\tau_q^{\pm1}]. \] We interpret the sheaf associated with a Weil divisor as a rank-one reflexive sheaf. Write $\tau_j$ for the canonical section of $D_j$. For a nonzero $f\in\C[C]$, its \emph{minimal homogeneous lift} is $\boldsymbol\tau^{\nu(f)}f$, where $\nu(f)_j=-\ord_{D_j}(f)$. It is not divisible by any boundary section. This agrees with the minimal monomial lifting convention of \cite{FranconeLifting,Francone}. \subsection{Reduced-word seeds} We choose a reduced expression $w=s_{i_1}\cdots s_{i_d}$. Let $p(k)$ and $s(k)$ be the preceding and following occurrences of $i_k$, with values $-\infty$ and $+\infty$ when absent. We write $w_{\leq k}=s_{i_1}\cdots s_{i_k}$ and use the \emph{unipotent minors} $f_k=D_{\Lambda_{i_k},w_{\leq k}\Lambda_{i_k}}$. A position is mutable precisely when $s(k)<+\infty$. With $A=(a_{ij})$ the Cartan matrix, our exchange matrix has mutable positions indexing its rows and all positions indexing its columns: \begin{equation}\label{eq:word-matrix} B_{kj}=\begin{cases} 1,&j=p(k),\\ -1,&j=s(k),\\ a_{i_j i_k},&j0. \] Assume an algebraic torus acts diagonally in every seed and preserves the bracket. We also assume that the positive grading is induced by a one-parameter subgroup of this torus. We require its infinitesimal eigenvalue vectors to span $\ker B$, and call the resulting diagonal fields \emph{grading derivations}. \begin{proposition}\label{prop:hamiltonian-recognition} Let $D$ be a nonzero locally nilpotent derivation of degree zero and nonzero torus weight. Let $b$ be frozen and suppose that $a=Db\ne0$. Assume that a grading derivation $\mathsf G$ and a nonzero real number $\eta_D$ satisfy \begin{equation}\label{eq:hamiltonian-recognition} bD=\eta_D\{a,-\}+a\mathsf G. \end{equation} If a seed satisfies $B_{kb}/\eta_D\geq0$ at every mutable position, then it is adapted to $D$. More precisely, \[ D=c_Dx^v\partial_b,\qquad c_D\in\C^\times,\qquad v_b=0,\qquad v_k=\frac{\delta_k}{\eta_D}B_{kb}\quad(k\text{ mutable}), \] and every entry of $v$ is a nonnegative integer. The same conclusion holds, without the column-sign assumption, if $Db$ is already a scalar multiple of a monomial in the seed coordinates other than $b$, or if $D$ fixes every seed coordinate other than $b$. \end{proposition} \begin{proof} The mixed Laurent ring of the seed equals the localization of $R$ at its mutable variables. None of these variables is divisible by the frozen prime $b$, since each is a unit in that Laurent ring and $b$ is not. Consequently $b\Lp(\Sigma)\cap R=bR$. For a locally nilpotent derivation, the order of the last nonzero derivative is additive on products, by the Leibniz formula. It decreases by one under $D$, so $b\nmid Db$. The constant term $a_0$ of $a$, considered as a polynomial in $b$ with mixed Laurent coefficients, is therefore nonzero. Write $\mathsf Gx_j=g_jx_j$. For each exponent $v$ occurring in $a_0$, evaluate \eqref{eq:hamiltonian-recognition} on $x_j$, reducing modulo $b$. At $j=b$, first divide by $b$. We obtain $\eta_Dv\mathsf P+g=e_b$. The difference $u$ of two such exponents satisfies $u\mathsf P=0$. Their torus weights agree, so $u$ annihilates $\ker B$ and has the form $u=cB$. Compatibility forces $c=0$. Thus $a_0=c_Dx^v$ is one monomial. Transposing the last equation and multiplying by $B$ gives the asserted mutable exponents. These are nonnegative by the column hypothesis, and the frozen exponents are nonnegative because the coefficients are polynomial. Hence $x^v\in R$ and $a-c_Dx^v\in bR$. Both terms have the positive degree of $b$, because $D$ has degree zero and the seed is homogeneous. Their difference is therefore a scalar multiple of $b$. Its torus weight differs from the weight of $b$, so this scalar is zero. Substitution in \eqref{eq:hamiltonian-recognition} shows that $D$ kills every other seed coordinate. Suppose instead that $a$ is already a monomial in the required coordinates. Applying \eqref{eq:hamiltonian-recognition} to $x_j\ne b$ expresses $bDx_j$ as a scalar multiple of $ax_j$. The latter has $b$-valuation zero unless its scalar vanishes, whereas the former has positive valuation. Adaptedness follows, and compatibility gives the same exponent formula. Finally, suppose that $D$ fixes every seed coordinate other than $b$. Applying \eqref{eq:hamiltonian-recognition} to the coordinates shows that every monomial of $a$ satisfies $\eta_Dv\mathsf P+g=e_b$. The same uniqueness argument makes $a$ a single Laurent monomial. Its exponent at $b$ is zero because $b\nmid Db$. Since $a\in R$, the Laurent-monomial assertion of \cite[Lemma~4.2]{FeiSchubert} makes every mutable exponent nonnegative as well. \end{proof} \begin{corollary}\label{cor:graded-covariance} Keep the assumptions on $D$ and $b$, and suppose that $D^2b=0$. If a grading derivation $A_D$ satisfies \[ \cL_D\Pi=D\wedge A_D,\qquad [A_D,D]=\lambda_DD,\qquad \lambda_D\in\mathbb R\setminus\{0\}, \] then \eqref{eq:hamiltonian-recognition} holds with $\eta_D=-\lambda_D^{-1}$. Thus the conclusions of Proposition~\ref{prop:hamiltonian-recognition} apply. \end{corollary} \begin{proof} Set $T=\{b,-\}/b$. Compatibility and the grading-span assumption make $T$ a grading derivation. Write $[T,D]=\mu D$ and $A_Db=hb$. Covariance evaluated on $(b,f)$ gives \[ \{a,f\}=a(T-A_D)f+b(h-\mu)Df. \] Since $Ta=\mu a$, $A_Da=(h+\lambda_D)a$, and $Da=0$, evaluation at $f=a$ yields $\mu=h+\lambda_D$. The desired identity follows with $\mathsf G=\lambda_D^{-1}(T-A_D)$. \end{proof} An element $f\in R$ is \emph{Poisson-normal} if $\{f,R\}\subseteq fR$. \begin{lemma}\label{lem:grading-twist} Let $H_1,\ldots,H_q$ be commuting grading derivations. Replacing $\Pi$ by \[ \Pi'=\Pi+\sum_{u1$, every degree in \eqref{eq:untwisted-origin-degrees} is positive. Since the minimal lift of an absent minor has degree zero, $w$ then has full affine support. \subsection{Source translation and the Poisson pencil} \begin{lemma}\label{lem:untwisted-framed-borel-cell} The open Schubert cell is the moduli space of framed $B_-$-bundles on $\mathbb P^1$ with torus degree $\lambda$, with the opposition convention fixed above. Source translation defines a regular additive-group action $\Psi_a$ on the cell. It commutes with the finite Cartan torus and preserves the polynomial volume form. \end{lemma} \begin{proof} For a positive finite root $\beta$, set $m_\beta=\langle\beta,\lambda\rangle$. A transition function has the form $n_-(u)u^{-\lambda}$. Changes of unipotent trivialization on the two affine charts, with value one at infinity, give the quotient \[ \C[u,u^{-1}]/\bigl(u^{-1}\C[u^{-1}]+u^{m_\beta}\C[u]\bigr) \] in root direction $-\beta$. Order the roots by height. Commutators involve only roots of greater height, so successive elimination gives the unique representative \[ \left(\prod_{\beta>0}x_{-\beta}\left(\sum_{0\le j1$ and use $v=s_rw$. Since $\pi$ preserves the affine marks, $j_*\ne0$. The affine translation formula gives \begin{equation}\label{eq:boundary-residue-root} v^{-1}(\delta_{\mathrm{aff}}-\alpha_r)=\alpha_{j_*}. \end{equation} \begin{proposition}\label{prop:boundary-residue-root-action} On $\eta\ne0$, the divisor $\tau=0$ is the product of the grading torus and the Schubert cell $U(v)$. The translation field has a decomposition $J=\cA+t^{-1}\cB$, where $\cA$ and $\cB$ act on $\C[U(v)][\eta^{\pm1}]$ and annihilate $t$ and $\eta$. Under Chevalley transposition \[ \iota_v:U(v)\longrightarrow U(v^{-1}), \qquad n\longmapsto\dot v^{-1}n^{\mathsf t}\dot v, \] the residue derivation $\cB$ is a nonzero scalar multiple of the positive simple-root operator $E_{j_*}$. \end{proposition} \begin{proof} We use a Gauss chart near the divisor. Choose a reduced word for $w$ beginning with $r$, and delete that first letter to obtain a word for $v$. The section $\eta$ is a Weyl translate of $\tau$, so $\eta\ne0$ is the corresponding translate of the open Schubert cell. The rank-one Gauss chart on $\eta\ne0$ has representatives \[ g(t,n;u)=x_{-r}(t)n(u)\dot s_r u^\lambda, \qquad n\in U(v),\quad \lambda=\omega_r^\vee. \] We choose the signs in $t$ and the pinned root homomorphisms so that the first positive coordinate on $\tau\ne0$ is $1/t$. The infinitesimal lowering action is translation in $t$. Consequently the $F_r$-invariant coordinates in this chart are precisely the coordinates of $U(v)$, together with $\eta^{\pm1}$. At $t=0$ this is the open cell of the boundary Schubert variety. The finite parts of the inversion roots of $v$ are positive and different from $\alpha_r$. Indeed, its inversion roots are obtained by applying $s_r$ to those of $w$ other than $\alpha_r$, and $s_r$ preserves every other positive finite root. We may therefore regard $n(u)$ as a polynomial loop in the unipotent radical generated by those roots. This radical is normalized by the finite rank-one subgroup corresponding to $r$. In the positive-Borel description of the framed bundle, with transition function $n_+(u)u^\lambda$, normalized source translation is generated by \[ Jg=-\partial_u g+u^{-1}\zeta g, \qquad \zeta=\Ad_{n_+(0)}\lambda. \] This follows by removing the unique negative Laurent term introduced by the change of source coordinate. Equivalently, on the positive unipotent factor it is \[ Jn_+=-n_+'+ u^{-1}\bigl(\Ad_{n_+(0)}\lambda\,n_+-n_+\lambda\bigr). \] Set $\lambda'=s_r\lambda=\lambda-\alpha_r^\vee$. The rank-one Gauss identity gives \begin{equation}\label{eq:boundary-framing-residue} \zeta=\Ad_{x_{-r}(t)n(0)} \bigl(\lambda'-t^{-1}e_r\bigr). \end{equation} To verify this formula, use the factorization in Lemma~\ref{lem:boundary-prefix-normalization}, with $z=t^{-1}$. For a positive finite root $\gamma\ne\alpha_r$, $\langle\gamma,\lambda'\rangle =\langle s_r\gamma,\lambda\rangle\ge0$. Thus the constant complementary factor has the form $q^{(z)}(0)=x_r(-t)q_0$, where $q_0$ is generated by roots of $\lambda'$-degree zero. These root groups commute with $x_r$: a root $\gamma+\alpha_r$ would have negative $\lambda'$-degree. In particular $q_0$ centralizes $\lambda'$. The same Gauss factorization gives \[ \zeta =\Ad_{x_{-r}(t)n(0)h_r(z)x_r(-t)}\lambda' =\Ad_{x_{-r}(t)n(0)}(\lambda'-tz^2e_r), \] which is \eqref{eq:boundary-framing-residue} because $tz^2=t^{-1}$. Substitute \eqref{eq:boundary-framing-residue} into the formula for $Jg$. In the $n$-coordinate, its term with a pole at $t=0$ is represented by $-u^{-1}\Ad_{n(0)}e_r\,n$. We may add $u^{-1}ne_r$, since this is tangent to the right stabilizer of the partial-flag point. In fact, \eqref{eq:boundary-residue-root} says that conjugating $e_r/u$ by $\dot v^{-1}$ gives a nonzero multiple of the finite root vector $f_{j_*}$. Since $j_*\ne q$, this vector belongs to the Levi subalgebra of the right parabolic. The resulting polynomial vector field is \begin{equation}\label{eq:boundary-dressing-field} \cB(n)=\frac{ne_r-\Ad_{n(0)}e_r\,n}{u}. \end{equation} The numerator has zero constant term. The regular part of the calculation is $-n'+u^{-1}(\Ad_{n(0)}\lambda'\,n-n\lambda')$. The first coordinate $\eta/\tau$ is the constant polynomial coordinate for the simple root $\alpha_r$. Source translation fixes it, and hence fixes both $t$ and $\eta$. This proves the stated decomposition. It remains to identify this vector field as a root action. Let $X=e_r/u$, a negative real affine root vector. The affine Chevalley anti-involution $\mathsf t$ sends $e_\alpha u^k$ to a fixed nonzero multiple of $f_\alpha u^{-k}$; in particular, it reverses the loop degree. Consider the Gauss factorization \begin{equation}\label{eq:boundary-right-gauss} n\exp(sX)=b_-(s)n_s, \qquad n_s\in U(v), \quad b_-(s)\in U_-\cap\dot vU_-\dot v^{-1}. \end{equation} This factorization is the transpose of the usual positive root action on $U(v^{-1})$. More explicitly, Chevalley transposition changes its left side into $\exp(s'e_{j_*})\iota_v(n)$, where $s'$ is a fixed nonzero multiple of $s$, by \eqref{eq:boundary-residue-root}. It changes the right side into $\iota_v(n_s)\iota_v(b_-(s))$, whose second factor belongs to $U_+\cap\dot v^{-1}U_+\dot v$. This is the complementary positive subgroup in the Schubert-cell factorization for $v^{-1}$. Hence \eqref{eq:boundary-right-gauss} exists on the whole cell and describes that positive simple-root action. Differentiating \eqref{eq:boundary-right-gauss} at $s=0$ gives $\dot n=nX-\dot b_-n$. The negative affine part of $\Ad_nX$ is exactly $u^{-1}\Ad_{n(0)}e_r$: all its finite root vectors are positive, and its Laurent expansion has no power below $u^{-1}$. Consequently $\dot b_-=u^{-1}\Ad_{n(0)}e_r$, and $\dot n$ is precisely \eqref{eq:boundary-dressing-field}. We verify the complementary subgroup in loop coordinates. For a negative-gauge root $\gamma+k\delta_{\mathrm{aff}}$, where $\gamma$ is finite positive and $k<0$, the root $v^{-1}(\gamma+k\delta_{\mathrm{aff}})$ is negative. For $\gamma\ne\alpha_r$, its current coefficient is $k-\langle\gamma,\lambda'\rangle<0$; for $\gamma=\alpha_r$, it has finite part $-\alpha_r$ and current coefficient $k+1\le0$. The length-zero factor, when present, preserves signs of affine roots. Thus every removed negative-gauge factor is carried by transposition to the required positive complement. This identifies the residue with the asserted simple-root action. \end{proof} \subsection{The translation field in two Gauss charts} On the chart $\eta\ne0$, Proposition~\ref{prop:boundary-residue-root-action} gives \[ J=\cA+t^{-1}\cB,\qquad Jt=J\eta=0, \] where $\cA,\cB$ preserve $\C[U(v)][\eta^{\pm1}]$ and annihilate $t,\eta$. Since $\cB$ is a root derivation, it is locally nilpotent. \begin{lemma}\label{lem:boundary-commuting-flows} The derivations $\cA$ and $\cB$ commute. Consequently $\cD=\tau J$ and $\cD^-=[F_r,\cD]$ are commuting regular locally nilpotent derivations on the Cox ring, of Cartan degree one. They annihilate $\tau$ and $\eta$. Under the adjoint action of the finite rank-one subgroup associated with $r$, they span its natural two-dimensional module. \end{lemma} \begin{proof} Choose a faithful matrix representation and expand $n(u)=n_0+un_1+\cdots$. Set \[ H=\Ad_{n_0}\lambda',\qquad X=\Ad_{n_0}e_r, \qquad Q=n_1n_0^{-1}. \] The normalized fields computed in the preceding proposition are \[ \cA n=-n'+\frac{Hn-n\lambda'}u, \qquad \cB n=\frac{ne_r-Xn}u. \] Taking their constant coefficients gives \[ (\cA n_0)n_0^{-1}=-Q+[H,Q], \qquad (\cB n_0)n_0^{-1}=[Q,X]. \] Since $[\lambda',e_r]=-e_r$, we have $[H,X]=-X$, and the Jacobi identity yields \[ \cA X+\cB H =[-Q+[H,Q],X]+[[Q,X],H]=0. \] Substituting these identities in the commutator gives \[ [\cA,\cB]n =-\frac{(\cA X+\cB H)n}{u} +\frac{([H,X]+X)n+n([e_r,\lambda']-e_r)}{u^2}=0. \] On $R[(\tau\eta)^{-1}]$, the derivations $J$ and $t^{-1}\cB$ are commuting locally nilpotent derivations. Their difference $\cA$ is therefore locally nilpotent. Since $\cA$ preserves $\C[U(v)][\eta^{\pm1}]$, its restriction is locally nilpotent as well. The expression on the $\eta$-chart shows that $J$ has at most a simple pole along $\tau=0$, and it is regular on $\tau\ne0$. The distinct primes $\tau,\eta$ have no common prime divisor, so these charts meet every prime divisor. Normality gives $\cD=\tau J\in\Der(R)$. It is locally nilpotent because $J\tau=0$ and $J$ is locally nilpotent on $R[\tau^{-1}]$. Here $F_r=\partial_t$, so $\cD=\eta(t\cA+\cB)$ and $\cD^-=\eta\cA$. These fields commute and annihilate $\tau,\eta$. The commutator definition makes $\cD^-$ regular on $R$, and its expression $\eta\cA$ makes it locally nilpotent. Source translation commutes with the finite positive Borel action. Thus $[E_r,\cD]=0$ and $[H_r,\cD]=\cD$. The derivations of a fixed degree in a finitely generated positively graded algebra form a finite-dimensional rational module: restrict them to a finite-dimensional homogeneous generating space. The field $J$ is nonzero: on the highest-root line $n_+(u)=x_\theta(z u^{M-1})$, its formula above changes the coefficient of $u^{M-2}$ at rate $z$. Thus $\cD\ne0$, and its highest weight-one string is $\cD,[F_r,\cD]$. In particular, $[E_r,\cD^-]=\cD$ and $[F_r,\cD^-]=0$, as required. \end{proof} The rank-one invariant $\Theta_r=\tau\cD^--\eta\cD$ is consequently a regular locally nilpotent derivation, and $\cB=-\Theta_r/\eta^2$. This gives a uniform way to calculate pole orders in every untwisted component. For a locally nilpotent derivation $D$ and a nonzero element $f$, we write $\deg_D f=\max\{j:D^jf\ne0\}$. \begin{proposition}\label{prop:translation-pole-degree} Let $x\in R\setminus\{0\}$ be homogeneous and satisfy $F_rx=0$. For every $a\ne0$, the pole order of $\Psi_a(x)$ along $\tau=0$ is $\deg_{\Theta_r}(x)$. Consequently a prefix minor in a reduced word beginning with $r$ satisfies \[ \nu_k(a)=\nu_k(0)+\deg_{\Theta_r}\bigl(x_k^{(0)}\bigr). \] \end{proposition} \begin{proof} By Lemma~\ref{lem:boundary-prefix-normalization}, $R[\eta^{-1}]=S[t]$, where $S=\C[U(v)][\eta^{\pm1}]$ and $F_r=\partial_t$. Thus $x\in S$. The preceding lemma gives commuting locally nilpotent derivations $\cA,\cB$ of $S$, with $J=\cA+t^{-1}\cB$ and $\cB=-\eta^{-2}\Theta_r$. It follows that \[ \Psi_a(x)=\exp(a\cA)\left(\sum_{m\ge0}\frac{a^m}{m!}t^{-m}\cB^m x\right). \] The sum is finite. Its coefficient at the most negative power of $t$ is nonzero, because $\exp(a\cA)$ is an automorphism of $S$ and $a\ne0$. Thus its pole order is exactly $\deg_{\cB}(x)$. Since $\cB\eta=0$, this equals $\deg_{\Theta_r}(x)$. At the generic point of $\tau=0$, the function $\eta$ is a unit. Hence $t=\tau/\eta$ and $\tau$ have the same valuation, which proves the first assertion. The second follows by applying it to the origin minimal lift, which is $F_r$-invariant. \end{proof} \subsection{The Hamiltonian of the boundary section} Let $K$ be the derivation of the dual current grading, so that a root coordinate of mode $j$ has degree $-j$. The framed normalization is homogeneous in $u$ and $a$, so $ [K,J]=-J $. The derivation $J$ commutes with the finite Cartan fields and fixes $\tau$. \begin{lemma}\label{lem:uniform-translated-nonpoisson} The canonical Cox pencil $\Pi_a$ satisfies \begin{equation}\label{eq:uniform-normal-Hamiltonian} \{\tau,-\}_a=\{\tau,-\}_0+\frac{a}{L}\cD, \qquad \cD=\tau J. \end{equation} If $\langle\theta,\lambda\rangle\ge2$, the divisor $\tau=0$ is not Poisson for $\Pi_a$ when $a\ne0$. \end{lemma} \begin{proof} Write $H_{\tau,0}=\tau^{-1}\{\tau,-\}_0$. For a terminal cell minor $f_i$, its logarithmic Hamiltonian has rotation coefficient $-k_i$. Indeed, \[ \{\log f_i,g\}_0 =-\tfrac12((w+1)\Lambda_i,\wt(g))g, \] and both $w\Lambda_i$ and $\Lambda_i$ have level $k_i$. The remaining part is a finite Cartan field. Equation~\eqref{eq:cell-volume} expresses the cell volume form as the product of the terminal minors times the logarithmic coordinate volume. The divergence formula for the canonical lift therefore gives $H_{\tau,0}|_{\C(Y)}=-\kappa^{-1}\sum_i\{\log f_i,-\}_0$. Since $\kappa=Lh^\vee$ and $\sum_i k_i=h^\vee$, the rotation coefficient of $H_{\tau,0}$ is $1/L$. Thus $H_{\tau,0}$ is $K/L$ plus finite Cartan and grading fields. The grading field is determined by $H_{\tau,0}\tau=0$. In particular $H_{\tau,0}$ is regular on the Cox ring. By Proposition~\ref{prop:untwisted-poisson-pencil}, $\Pi_a=\Psi_a\Pi_0$. Since $\Psi_a\tau=\tau$, we obtain $H_{\tau,a}=\Psi_aH_{\tau,0}\Psi_a^{-1}$. The commutator $[J,H_{\tau,0}]=J/L$ has no further nonzero commutator with $J$. Consequently $H_{\tau,a}=H_{\tau,0}+aJ/L$, which proves \eqref{eq:uniform-normal-Hamiltonian}. We check nonvanishing on the boundary. Let $v=s_rw$. The index $j_*$ belongs to the support of $v$. Indeed, otherwise $v\Lambda_{j_*}=\Lambda_{j_*}$. The finite part of the left side, obtained from $w=t_{-\lambda}\pi$, is $(1-M)(\omega_r-\alpha_r)$. For any finite node adjacent to $r$, its pairing with that simple coroot is negative, whereas $\omega_{j_*}$ is dominant. The positive simple-root operator $E_{j_*}$ on $U(v^{-1})$ is therefore nonzero. The residue identification makes $\cB\ne0$. In the $\eta$-chart, $\cD=\eta(t\cA+\cB)$, and hence $\cD\bmod\tau=\eta \cB\ne0$. The first term on the right of \eqref{eq:uniform-normal-Hamiltonian} is divisible by $\tau$. The second has nonzero residue when $a\ne0$, proving that $(\tau)$ is not a Poisson ideal. \end{proof} \begin{corollary}\label{cor:terminal-degrees-from-residue} Assume $\langle\theta,\lambda\rangle\ge2$. For $a\ne0$, the minimal lift of each translated terminal minor has degree $Lk_i$. More precisely, if $\widetilde b_i$ is its origin minimal lift, then $\Psi_a(\widetilde b_i)$ is regular for $i\ne j_*$, whereas $\tau\Psi_a(\widetilde b_{j_*})$ is the minimal lift of the terminal $j_*$-minor. \end{corollary} \begin{proof} After division by powers of $\eta$, the origin terminal lifts are precisely the terminal minors of $U(v)$. Affine Chevalley transposition preserves their simple-root indices. On $U(v^{-1})$, the positive simple-root operator $E_{j_*}$ annihilates each terminal $i$-minor for $i\ne j_*$, while its string on the terminal $j_*$-minor has length one. Since $\cB\eta=0$, it follows that $\deg_{\cB}(\widetilde b_i)=\delta_{i,j_*}$. The exact pole-degree formula for $J=\cA+t^{-1}\cB$ gives $\nu_i(a)=\nu_i(0)+\delta_{i,j_*}$. At the origin the boundary coroot gives $\nu_i(0)=Lk_i-\delta_{i,j_*}$, so $\nu_i(a)=Lk_i$, as claimed. \end{proof} \subsection{Recovering an exchange from a divisor} The next theorem recovers an exchange relation from a prime divisor that is not Poisson. It allows us to \emph{unfreeze} a coordinate, that is, to make it mutable, by constructing its exchange relation. \begin{theorem}\label{thm:poisson-unfreezing} Let $R$ be a finitely generated normal complex domain with a Poisson bracket, and let $x_0,x_1,\ldots,x_{N-1}$ be pairwise nonassociated prime elements which generate its fraction field and form a transcendence basis. Suppose \[ \{x_i,x_j\}=p_{ij}x_ix_j,\qquad \mathsf P=(p_{ij})\in\Mat_N(\mathbb Q). \] Let an algebraic torus act diagonally on these elements and by Poisson automorphisms on $R$, and let $\mathsf W$ be the matrix of their characters. Assume that \[ v\mathsf P=0,\quad v\mathsf W=0\quad\Longrightarrow\quad v=0. \] Assume also that the prime ideal $(x_0)$ is not a Poisson ideal. Then there is a primitive vector $v=(0,v_1,\ldots,v_{N-1})\in\Z^N$, unique up to sign, and a constant $c\in\C^\times$ such that \[ x_0'=\frac{\prod_{j>0}x_j^{[v_j]_+}-c\prod_{j>0}x_j^{[-v_j]_+}}{x_0}\in R. \] The numerator vanishes to order exactly one along $(x_0)$. Furthermore $v\mathsf W=0$ and $v\mathsf P=-\gamma e_0$ for a nonzero rational number $\gamma$. Suppose a compatible extended exchange matrix $B$ has its mutable indices among $1,\ldots,N-1$, with $B_i\mathsf P=-\delta_i^{-1}e_i$ and $\delta_i>0$. Choose the sign of $v$ so that $\gamma>0$. Adjoining $v$ as the exchange row at $0$ produces a compatible matrix of one greater rank, with primitive new row and skew-symmetrizable principal part. After scalar normalization, $x_0'$ is its first exchange variable. \end{theorem} \begin{proof} Write $D=V(x_0)$ and let $\mathscr V=R_{(x_0)}$ be its discrete valuation ring. Every $x_j$ with $j>0$ is a unit in $\mathscr V$. The residue homomorphism \[ A=\C[x_1^{\pm1},\ldots,x_{N-1}^{\pm1}]\longrightarrow\C(D) \] has a prime kernel $I$. For $j>0$, the Hamiltonian derivation $x_j^{-1}\{x_j,-\}$ preserves $\mathscr V$ and its maximal ideal, since $\{x_j,x_0\}/x_j=-p_{0j}x_0$. The geometric torus has the same property. The restrictions of these derivations to $A$ are diagonal. Since $\mathsf P$ is rational and $\mathsf W$ is integral, they generate the Lie algebra of a diagonal subtorus. The ideal $I$ is invariant under this torus. The character lattice of the quotient by this diagonal torus consists of the integer vectors $u\in\Z^{N-1}$ satisfying $u\mathsf P_{>0,>0}=0$ and $u\mathsf W_{>0}=0$. For such a vector, $v=(0,u)$ satisfies $v\mathsf P=-\gamma e_0$ and $v\mathsf W=0$. Our nondegeneracy hypothesis makes the map $u\mapsto\gamma$ injective. Thus the quotient torus has dimension at most one. We claim that $I\ne0$. Otherwise the contraction of the maximal ideal of $\mathscr V$ to $\C[x_0,x_1^{\pm1},\ldots,x_{N-1}^{\pm1}]$ would be $(x_0)$. Its localization at this prime and $\mathscr V$ are discrete valuation rings in the same fraction field, and one dominates the other; hence they are equal. The log-canonical formula makes $(x_0)$ a Poisson ideal in this valuation ring. Therefore $\{x_0,R\}\subset x_0\mathscr V\cap R=x_0R$, contrary to the assumption. An invariant proper prime ideal in a Laurent polynomial algebra whose quotient torus has dimension at most one is generated by $x^u-c$ for a primitive nonzero quotient character $u$ and a nonzero constant $c$. Indeed the diagonal torus acts transitively if the quotient has dimension zero; in dimension one its proper irreducible invariant subvarieties are the fibers of the quotient character. This proves the asserted binomial relation. Clearing its Laurent monomial denominator gives a polynomial $F$ divisible by $x_0$ in $R$, so $x_0'=F/x_0$ belongs to $R$. The scalar $\gamma$ is nonzero by nondegeneracy. If the logarithmic weights of the two monomials of $F$ under $\{x_0,-\}/x_0$ are $a_+$ and $a_-$, then $a_+-a_-=\gamma\ne0$. Their residues agree up to $c$ and are nonzero, so $\{x_0,F\}/x_0$ has a nonzero residue along $D$. Thus $\ord _D\{x_0,F\}=1$. If $F=x_0^m h$ in $\mathscr V$, with $h$ a unit, then $\{x_0,F\}=x_0^m\{x_0,h\}$ and the latter factor is in $\mathscr V$. Therefore $m\le1$, and divisibility gives $m=1$. For an existing mutable index $i$, skew-symmetry of $\mathsf P$ gives \[ -\gamma B_{i0}=v\mathsf P B_i^{\mathsf T}=-v(B_i\mathsf P)^{\mathsf T}=\delta_i^{-1}v_i. \] Thus $\delta_0=\gamma^{-1}$ extends the symmetrizer to the new mutable index. The new exchange vector is independent of the old ones, because its product with $\mathsf P$ is supported at $0$. Its primitivity follows from that of $u$. Scalar normalization changes the nonzero coefficient $-c$ to the usual plus sign; simultaneous normalization with existing exchanges uses the full row rank of the augmented matrix. \end{proof} To obtain an upper cluster structure from this exchange, we apply Proposition~\ref{prop:volume-criterion}. It remains to check homogeneity, regularity of the other first exchanges, and the equality between the sum of the frozen degrees and the degree of a homogeneous volume form. \subsection{The translated upper cluster structure} \begin{theorem}\label{thm:untwisted-translated-cluster} For every untwisted KR Cartan algebra and every $a\ne0$, there is a seed $\Sigma_a$ satisfying \[ R=\Up(\Sigma_a). \] It has $d-n$ mutable variables and $n+1$ non-inverted frozen variables, represented up to nonzero constants by the geometric sections \[ b_i(a)=\tau^{Lk_i}\Psi_a(f_i),\qquad 0\le i\le n. \] The seed is homogeneous for Cartan degree and finite weight, the extended exchange matrix has full rank and primitive rows, and $\Pi_a$ is compatible with it. The variables $b_i(a)$ are pairwise nonassociated primes of degrees $Lk_i$. For $M>1$, the seed is obtained by minimally lifting the translated cell seed and making $\tau$ mutable. For $M=1$, it is the origin seed, independent of $a$. \end{theorem} \begin{proof} If $M=1$, source translation is trivial. Formula \eqref{eq:untwisted-origin-degrees} gives zero precisely at the missing index $j_*$, and the other terminal degrees are $Lk_i$. Thus $b_{j_*}(a)=\tau$, and the other $b_i(a)$ are the terminal lifts. Adjoining $\tau$ gives the stated $n+1$ frozen variables and $d-n$ mutable variables. The assertions follow from Corollary~\ref{cor:kr-origin-cluster}. We now assume $M>1$. We choose reduced-word cell coordinates $g_1,\ldots,g_d$ and write $g_j^{(a)}=\Psi_a(g_j)$. Translation preserves their primality and the regularity of their first exchanges on the cell. Write $\nu_j=-\ord_{Y\setminus C}(g_j^{(a)})$. Each $\nu_j$ is positive: otherwise the nonconstant cell function would extend to the projective normal variety $Y$. Its minimal lift $x_j=\tau^{\nu_j}g_j^{(a)}$ is therefore a prime element of $R$, coprime to $\tau$. The coordinates $x_1,\ldots,x_d,\tau$ are pairwise nonassociated and generate the fraction field. Homogenize the cell exchanges by the $\tau$-column. The old mutable rows keep their successor entry of absolute value one, so their rank and primitivity are unchanged. Their exchange quotients are regular after inverting $\tau$, and their denominators are primes coprime to $\tau$. They consequently belong to $R$. The volume identity \eqref{eq:cell-volume} is preserved by source translation. Lemma~ \ref{lem:canonical-lift-log-canonical} therefore proves log-canonicity and compatibility of the homogeneous lift. We verify the nondegeneracy hypothesis in Theorem~ \ref{thm:poisson-unfreezing}. In ordered root coordinates, let $\beta_1,\ldots,\beta_d$ be the inversion roots. Up to a common nonzero scalar, the leading skew matrix is $\mathsf P^{\mathrm{root}}_{jk}=(\beta_j,\beta_k)$ for $j1$, the origin seed has $d-n-1$ mutable variables and the additional frozen variable $\tau$. At $a=0$, the sections above satisfy $b_i(0)=\tau^{\delta_{i,j_*}}\widetilde b_i$, by \eqref{eq:untwisted-origin-degrees}. \section{Uniform charts for the shifted current operators} \label{sec:uniform-currents} We continue with an untwisted KR Cartan algebra. For $a\ne0$, the finite jet quotient defines the shifted affine generators \[ E_0(a)=f_\theta\otimes(u+a),\qquad F_0(a)=e_\theta\otimes(u+a)^{-1},\qquad H_0=-h_\theta. \] Here the inverse is expanded in $\C[u]/(u^M)$, where $M=\langle\theta,\lambda\rangle$. For finite nodes we set $E_i(a)=E_i$ and $F_i(a)=F_i$, and we set $t_0=1$. These operators satisfy the Chevalley relations. They are locally nilpotent: each preserves the finite-dimensional Cartan-degree spaces and has nonzero finite root weight. \begin{theorem}\label{thm:untwisted-uniform-currents} For every untwisted affine type and every $a\ne0$, the mutation class of $\Sigma_a$ contains an adapted seed for every finite Chevalley operator and for $F_0(a)$. The active frozen variable is $b_i(a)$ for finite $E_i$, and $b_{\pi^{-1}(i)}(a)$ for every $F_i$, including $i=0$. The mutation class also has a chart adapted to $E_0(a)$, with active frozen variable $b_0(a)$. In each of these charts the operator is a nonzero scalar times a monomial in the other coordinates multiplied by the derivative in its active frozen variable. Its mutable exponents are \begin{equation}\label{eq:shifted-current-exponents} m_k(E_i)=\frac{\delta_k}{t_i}B^{E_i}_{k,b_i}, \qquad m_k(F_i)=-\frac{\delta_k}{t_i}B^{F_i}_{k,b_{\pi^{-1}(i)}}. \end{equation} When $M>1$, the newly mutable boundary position has $\delta_\tau=L$. In these cases, a chart for $F_r$ is obtained by one final mutation at $\tau$; its coefficient is $\tau'$ or $\eta\tau'$, up to a nonzero scalar, where $\eta=F_r\tau$. \end{theorem} We construct the finite root charts first, followed by the charts for $F_0(a)$ and $E_0(a)$. \subsection{The boundary sections and finite root strings} We first study the component section $c=\tau^Lf_q$, where $q=\pi^{-1}(0)$. \begin{lemma}\label{lem:component-section-invariant} Source translation fixes $c$. If we set $\omega_0=0$, the translated boundary sections have finite weights \[ \wt b_i(a)=\omega_{\pi(i)}-\omega_i. \] Their weights for $H_0=-h_\theta,H_1,\ldots,H_n$ are $\Lambda_{\pi(i)}-\Lambda_i$, interpreted at level zero. \end{lemma} \begin{proof} Translation is trivial in the evaluation cases, so first suppose $M>1$. Then $q\ne j_*$. The origin section $c=\widetilde b_q$ is $F_r$-invariant. After division by its power of $\eta$, it is the terminal $q$-minor on the shortened cell. The residue derivation $\cB$ is the positive root operator $E_{j_*}$ on $U(v^{-1})$, and therefore $\cB c=0$. In the $\eta$-chart this gives $Jc=\cA c$, which is regular. It is regular on $\tau\ne0$ as well. Since these two open sets meet every prime divisor, normality gives $Jc\in R_L$. Under the untwisted KR--Demazure identification, $R_L$ is the dual of the Demazure submodule in the level-one module $V_{\mathrm{aff}}(\Lambda_q)$, with the constant shift of current degree that gives the cyclic vector degree zero. Write $p_\xi$ for an extremal coordinate section. Since $p_{w\Lambda_q}=\tau^L$ and $f_q=p_{\Lambda_q}/p_{w\Lambda_q}$, after normalization we have $c=p_{\Lambda_q}$. Thus $c$ is the coordinate functional on the affine highest-weight line. In the affine derivation grading, every weight of $V_{\mathrm{aff}}(\Lambda_q)$ has degree at most that of its highest vector. Consequently $c$ lies in the lowest dual current degree of $R_L$. The identity $[K,J]=-J$ would place a nonzero $Jc$ in the next lower degree. Thus $Jc=0$, and exponentiation proves the first claim. The finite weight of the cell minor $f_i$ is $\omega_{\pi(i)}-\omega_i-Lk_i\omega_r$. Translation commutes with the finite Cartan torus, and $\tau$ has weight $\omega_r$. Its factor $\tau^{Lk_i}$ gives the stated weight of $b_i(a)$. The same formula holds for the missing index in an evaluation case, where $b_{j_*}(a)=\tau$. Finally $\pi$ preserves the comarks. Pairing the finite weight with $-\theta^\vee$ gives $\delta_{0,\pi(i)}-\delta_{0,i}$; the finite simple-coroot pairings are immediate. These are precisely the asserted affine Cartan weights. \end{proof} \begin{lemma}\label{lem:finite-boundary-strings} For every finite node $j$, \[ \begin{gathered} E_j b_i(a)=0\quad(i\ne j),\qquad F_j b_i(a)=0\quad(i\ne\pi^{-1}(j)),\\ E_j^2b_j(a)=F_j^2b_{\pi^{-1}(j)}(a)=0. \end{gathered} \] Moreover, $E_jb_j(a)$ and $F_jb_{\pi^{-1}(j)}(a)$ are nonzero. \end{lemma} \begin{proof} At the origin these statements follow from the fundamental root strings and the adapted charts of Corollary~\ref{cor:kr-origin-currents}. Source translation commutes with every finite positive root subgroup and with the finite Levi subgroup generated by the simple roots other than $r$. These actions fix $\tau$, so the statements for all $E_j$ and for $F_j$, $j\ne r$, pass to the translated homogeneous lifts. The evaluation cases require no further argument. Suppose $M>1$, and write $b_* =\widetilde b_{j_*}$. On the $\eta$-chart, $F_r=\partial_t$, and the commuting fields $\cA,\cB$ annihilate $t,\eta$. The root string on $U(v^{-1})$ gives $\cB^2b_*=0$ and $\cB b_*\ne0$. The translated section is therefore \[ b_{j_*}(a)=\tau\exp(a\cA)b_*+a\eta\exp(a\cA)(\cB b_*). \] It follows that $F_r b_{j_*}(a)=\eta\exp(a\cA)b_*\ne0$ and $F_r^2b_{j_*}(a)=0$. For $i\ne j_*$, the residue kills $\widetilde b_i$, and $b_i(a)=\exp(a\cA)\widetilde b_i$ is independent of $t$. This proves the remaining annihilations. \end{proof} \begin{lemma}\label{lem:shifted-current-covariance} Let $\Pi_a$ be the canonical Cox lift of the translated trigonometric Poisson structure, and let $\mathsf E$ be the Cartan-degree derivation. For every shifted simple root operator $D=E_i(a)$ or $F_i(a)$, there is a grading derivation $A_D$ such that \[ \cL_D \Pi_a=D\wedge A_D,\qquad [A_D,D]=\begin{cases}-D/t_i,&D=E_i(a),\\ D/t_i,&D=F_i(a). \end{cases} \] If a frozen section $b$ satisfies $Db\ne0$ and $D^2b=0$, the Hamiltonian recognition identity has coefficient $t_i$ for $E_i(a)$ and $-t_i$ for $F_i(a)$. \end{lemma} \begin{proof} We use the same normalization as in the projective calculation. Let $\Omega_\g$ be the Casimir tensor for the invariant form with long roots of squared length two, and set $t_\alpha=2/(\alpha,\alpha)$ for every finite root $\alpha$. The trigonometric kernel centered at $-a$ is \[ \mathcal R_a(u,v)=\frac{u+v+2a}{2(u-v)}\Omega_\g +\frac12\sum_{\alpha>0}t_\alpha^{-1} (f_\alpha\otimes e_\alpha-e_\alpha\otimes f_\alpha). \] The coboundary description in \cite[Corollary~6.6 and Equation~(85)]{AbedinBurban}, after rescaling the invariant form to our normalization, identifies its commutator with $X(u)\otimes1+1\otimes X(v)$ as the loop cobracket. The simple-root relations give $H_i\wedge X/(2t_i)$ for both signs. For the affine simple root, we use $E_0(a)=f_\theta(u+a)$, $F_0(a)=e_\theta/(u+a)$, $H_0=-h_\theta$, and $t_\theta=1$. The matrix-coefficient bracket descends through the based gauge quotient, so the projective covariance is $\cL_{\overline D}\overline\Pi_a =\overline D\wedge(-\overline H_i/(2t_i))$, where $\overline\Pi_a$ is the projective tensor and bars denote projected vector fields. The current root fields preserve the homogeneous volume form, and the Cartan fields do so as their commutators. Proposition~\ref{prop:canonical-root-covariance} therefore lifts this identity with an Euler correction. If the volume degree is $\kappa$, the corrected field is $A_D=-H_i/(2t_i)-\lambda_D\mathsf E/\kappa$, where $\lambda_D=-1/t_i$ for a raising operator and $\lambda_D=1/t_i$ for a lowering operator. Since $[\mathsf E,D]=0$, its commutator eigenvalue is unchanged. A grading twist adds to $A_D$ a grading field commuting with $D$. Corollary~\ref{cor:graded-covariance} gives the asserted Hamiltonian coefficient whenever $Db\ne0$ and $D^2b=0$. \end{proof} \subsection{The finite root charts} Write $b_* =\widetilde b_{j_*}$ for the origin terminal lift. \begin{lemma}\label{lem:boundary-chart-completion} Suppose a sequence of origin mutations avoiding $\eta$ reaches a seed such that every coordinate other than $\tau$ is $F_r$-invariant and, on their fraction field, $\cB=Q\partial_{b_*}$. Assume that $Q$ is a Laurent monomial whose only possible negative exponent is at $\eta$. In the translated cluster structure, perform the same mutations and then mutate $\tau$. The resulting seed is adapted to $F_r$, with active frozen coordinate $b_{j_*}(a)$. Up to nonzero scalar normalizations, its coefficient is either $\tau'$ or $\eta\tau'$, where $\tau'$ is the new exchange variable. \end{lemma} \begin{proof} We first compare the coordinates along the two mutation paths. After inverting $\tau$, divide each variable by the power of $\tau$ equal to its Cartan degree. Both paths then give the same cell exchanges, related by $\Psi_a$. Therefore a variable along the translated path differs from the translated origin variable by a power of $\tau$. Every reachable cluster variable is prime in the factorial upper cluster algebra, so that power is uniquely determined by the requirement that its $\tau$-valuation be zero. Let $y\ne b_*,\tau$ be a coordinate of the final origin seed. Since $\cB y=0$, Proposition~\ref{prop:translation-pole-degree} shows that $y_a=\Psi_a(y)=\exp(a\cA)y$ is regular. It is $F_r$-invariant, so it cannot be divisible by $\tau$: the kernel of a locally nilpotent derivation is factorially closed, whereas $F_r\tau=\eta\ne0$. Thus the corresponding translated coordinates are the functions $y_a$. The active frozen coordinate is \begin{equation}\label{eq:completed-boundary-section} b_{j_*}(a)=\tau\exp(a\cA)b_*+a\eta\exp(a\cA)Q. \end{equation} Write $Q=\eta^mQ_0$, with $Q_0$ a monomial in the other coordinates of that seed, excluding $b_*,\tau,\eta$. The derivation $\Theta_r=-\eta^2\cB$ is regular. Therefore $\eta^{m+2}Q_0=-\Theta_r b_*$ belongs to the Cox ring. These coordinates are nonassociated primes, so $m\ge-2$. Let $p=\max(0,-m-1)$; thus $p$ is either zero or one. Modulo $\tau$, formula \eqref{eq:completed-boundary-section} gives the binomial relation \begin{equation}\label{eq:completed-boundary-exchange} \eta^p b_{j_*}(a)-a\eta^{p+1}\exp(a\cA)Q=0. \end{equation} Its two monomials are relatively prime. Indeed, if $p=1$ then $m=-2$, and the second monomial has no $\eta$-factor; if $p=0$, the first monomial is just $b_{j_*}(a)$. The slice decomposition $R[\eta^{-1}]=S[t]$ identifies the divisor $\tau=0$ on this chart with $\Spec S$. The remaining coordinates are algebraically independent in $S$: they are obtained from independent origin coordinates by the automorphism $\exp(a\cA)$. Equation~\eqref{eq:completed-boundary-exchange} expresses $b_{j_*}(a)$ as a Laurent monomial in them. It therefore generates the kernel of restriction from the Laurent ring in these coordinates and $b_{j_*}(a)$ to the generic $\tau$-divisor. Its exponent vector is primitive, because its $b_{j_*}(a)$-entry has absolute value one. The exchange vector at $\tau$ remains primitive under mutations at other indices. Indeed, the mutation formula preserves the ideal of $\Z$ generated by its entries: one inclusion follows from the formula, and the other follows from the inverse mutation. The numerator of the existing $\tau$-exchange vanishes on this divisor. Its exponent difference is a nonzero integral multiple of the primitive exponent difference in \eqref{eq:completed-boundary-exchange}. Primitivity makes this multiple $\pm1$; coprimality of its two monomials then identifies the exchange numerator up to a nonzero scalar. The new variable is therefore $\tau'=\eta^p\exp(a\cA)b_*$, whose factors are $F_r$-invariant. Applying $F_r$ to \eqref{eq:completed-boundary-section} gives \[ F_r b_{j_*}(a)=\eta\exp(a\cA)b_* =\eta^{1-p}\tau'. \] All other coordinates are fixed. Since $p\in\{0,1\}$, this is a monomial and proves the assertion. \end{proof} \begin{corollary}\label{cor:framing-symmetrizer} The new mutable position $\tau$ has symmetrizer $\delta_\tau=t_r=L$. In the completed lowering chart, the coefficient of $F_r$ involves only mutable variables. \end{corollary} \begin{proof} The letter $r$ occurs at least twice in the reduced word for $w$. Indeed, the first lift $\eta$ has degree one, whereas the minimal lift of the terminal $r$-minor has degree $M-\delta_{r,j_*}$. If $r\ne j_*$, this is at least two. If $r=j_*$, the residue calculation already shows that $r$ occurs in the shortened word. Thus $\eta$ is a mutable variable of the full cluster structure. It was frozen only to describe a restricted mutation path. The variable $\tau'$ is mutable as well, and Lemma~\ref{lem:boundary-chart-completion} gives the asserted coefficient. The exchange at $\tau$ has exponent one at $b_{j_*}(a)$, so the corresponding matrix entry has absolute value one. After mutation, the exponent of $\tau'$ in $F_r b_{j_*}(a)$ is also one. Hamiltonian recognition with coefficient $-t_r$ therefore gives $\delta_{\tau'}=t_r$. The symmetrizer is unchanged by mutation, proving the assertion for $\tau$. \end{proof} \begin{proposition}\label{prop:translated-finite-roots} The mutation class of $\Sigma_a$ contains adapted charts for every finite $E_i,F_i$. When $M>1$, the chart for $F_r$ is obtained by CGL reversal on the shortened cell, mutations at the nonterminal occurrences of $j_*$, and one final mutation at $\tau$. \end{proposition} \begin{proof} If $M=1$, translation is trivial and the origin finite-root corollary gives every required chart. We therefore assume $M>1$. Source translation commutes with the positive Borel action on the framed bundle. It also commutes with the finite Levi subgroup generated by the simple roots other than $r$. These operators fix $\tau$. We may therefore translate their origin adapted charts and take minimal homogeneous lifts. Homogenizing each exchange gives the same mutation path, with no mutation at $\tau$. Primality of the reachable variables ensures that the homogeneous lifts are minimal. Every coordinate other than the active frozen one is still invariant. The strings in Lemma~\ref{lem:finite-boundary-strings} and Hamiltonian recognition give the required adapted charts for $E_i$ and for $F_i$, $i\ne r$. For $F_r$, freeze $\eta$ and divide the remaining coordinates by the powers of $\eta$ prescribed by their Cartan degrees. Lemma~ \ref{lem:boundary-prefix-normalization} identifies this seed with the word seed of $U(v)$, where $v=s_rw$. CGL reversal reaches the seed of $U(v^{-1})$ for the reversed word, with the terminal minor indices unchanged. There the residue derivation is the positive root operator at $j_*$. Mutating its nonterminal occurrences gives a chart adapted to that derivation. Lift these mutations by $\eta$-homogeneity. Lemma~ \ref{lem:boundary-chart-completion} supplies the final mutation at $\tau$ and the asserted $F_r$-chart. \end{proof} \subsection{The highest-root current and source translation} For $0\le j0$. \begin{lemma}\label{lem:central-current-translation} On the Schubert cell, and on $R[\tau^{-1}]$, we have \[ [J,D_0]=0,\qquad [J,D_j]=-(M-j)D_{j-1}\quad(1\le j1$. \end{proposition} \begin{proof} Write $w'=ws_q$. Since $q$ is the only right descent of $w$, every reduced word for $w$ ends in $q$, and its final inversion root is $\beta=w'\alpha_q=-w\alpha_q =\theta+(M-1)\delta_{\mathrm{aff}}$. This root subgroup is central in the positive-current representative of the cell: $\theta+\alpha$ is not a finite root for any positive finite root $\alpha$. Thus its left and right actions agree. On the coordinate algebra, $D_{M-1}$ is a nonzero scalar multiple of differentiation along $n\exp(s e_\beta)$; the scalar accounts for the root-vector and dual-action conventions. The $q$-string through the highest vector of $V_{\mathrm{aff}}(\Lambda_q)$ has length one. Conjugating this string by a representative of $w'$ gives, for suitably normalized extremal vectors, \[ \exp(s e_\beta)v_{w\Lambda_q}=v_{w\Lambda_q}+\gamma s\,v_{w'\Lambda_q},\qquad e_\beta v_{w'\Lambda_q}=0,\qquad \gamma\ne0. \] The matrix-coefficient definition of the unipotent minor therefore gives $D_{M-1}f_q=\gamma D_{\Lambda_q,w'\Lambda_q}$, after absorbing the action scalar into $\gamma$. Let $p$ be the preceding occurrence of $q$. Every intervening letter fixes $\Lambda_q$, so $w'\Lambda_q=w_{\leq p}\Lambda_q$. Thus the derivative is the prefix minor $\gamma y_p$, where $y_p=D_{\Lambda_q,w_{\leq p}\Lambda_q}$. If there is no preceding occurrence, $w'\Lambda_q=\Lambda_q$ and we set $y_p=1$. Since $c=\tau^Lf_q$ and $D_{M-1}\tau=0$, we obtain $m_{M-1}=\gamma\tau^Ly_p$. If $x$ is the minimal Cox lift of $y_p$, of degree $\nu$, regularity of $m_{M-1}$ gives $m_{M-1}=\gamma\tau^{L-\nu}x$, with $L-\nu\geq0$. When $y_p=1$, take $x=1$ and $\nu=0$. The nonconstant prefix variable is mutable. For $M>1$, the translated derivative $\Psi_a(m_{M-1})$ is a Laurent monomial in $\tau$ and the translated prefix variable. Lemma~\ref{lem:central-current-translation} identifies it, up to a nonzero scalar, with the regular coefficient $F_0(a)c$. The translated prefix variable, when present, is prime and is not associated to $\tau$. Regularity therefore forces the exponent of $\tau$ to be nonnegative. Thus the coefficient is a monomial, and both possible factors are mutable. In an evaluation case, $M=1$ and source translation is trivial. The same calculation expresses $F_0(a)c=a^{-1}m_0$ as a monomial in the origin seed. Here $L=1$, and a factor of $\tau$ is allowed. The same root string gives $D_{M-1}^2c=0$. Lemma~\ref{lem:central-current-translation} also gives the operator identity \[ F_0(a)=(-1)^{M-1}a^{-M}\Psi_aD_{M-1}\Psi_a^{-1}. \] Since $\Psi_ac=c$, it follows that $F_0(a)^2c=0$. Shifted covariance and Hamiltonian recognition therefore make the initial translated seed adapted to $F_0(a)$. \end{proof} \subsection{The positive affine current} \begin{proposition}\label{prop:positive-affine-from-first-mode} Let $R$ be an untwisted KR Cartan algebra in the neutral component, and let $a\ne0$. Assume that the translated Schubert cluster structure has reachable adapted charts for every finite simple lowering operator. Then it also has a reachable adapted chart for $E_0(a)$. Its coefficient $E_0(a)c$ is a monomial in the variables of one seed. \end{proposition} \begin{proof} Write $m^-=(f_\theta\otimes u)c$ and $m^+=(e_\theta\otimes u)c$. By Corollary~\ref{cor:kr-origin-currents}, $m^-=E_0(0)c$ is a monomial in an origin seed. Since $c$ is finite $G$-invariant, the map $x\mapsto(x\otimes u)c$ is a homomorphism from the adjoint representation to $R_L$. It is nonzero because $m^-\ne0$, hence injective. Its two extremal vectors are $m^-$ and $m^+$. Every finite positive simple root has an adapted chart at the origin. Successive maximal raising derivatives carry $m^-$ to a nonzero scalar multiple of $m^+$. Proposition~ \ref{prop:maximal-root-derivative} shows that $m^+$ is a monomial in the variables of a seed in the origin mutation class. These mutations do not involve $\tau$, which is frozen there. The source-translation formula in current degree one gives $\Psi_a(m^+)=m^+$. Translate the preceding mutation path and homogenize its exchanges. The resulting path is reachable in the translated seed because it never mutates $\tau$. Its expression for $m^+$ is a Laurent monomial; the only possible new negative exponent is at $\tau$. Every other factor is a prime not associated to $\tau$, so regularity of $m^+$ forces this exponent to be nonnegative. Thus $m^+$ is again a monomial in seed variables. The assumed finite lowering charts and Proposition~\ref{prop:maximal-root-derivative} carry $m^+$ back to $m^-$, within the translated mutation class. Finite $G$-invariance of $c$ gives $E_0(a)c=(f_\theta\otimes(u+a))c=m^-$. At the origin, the fundamental affine root string gives $(f_\theta\otimes u)^2c=0$. Since $f_\theta c=0$ and $[f_\theta,f_\theta\otimes u]=0$, expanding the square gives $E_0(a)^2c=0$. Shifted root covariance supplies the Hamiltonian recognition identity. The monomial form of $E_0(a)c$ therefore makes this seed adapted.\qedhere \end{proof} \begin{proposition}\label{prop:evaluation-positive-affine} In every evaluation case, the origin mutation class contains a seed adapted to $E_0(a)=a f_\theta$. \end{proposition} \begin{proof} We have $M=L=1$. The sections $b_0=\tau$ and $c=b_q$ are respectively the highest and lowest vectors of the irreducible finite module $R_1$, in our opposition convention. The highest-root string through either vector has length one. Hence $e_\theta c$ and $f_\theta b_0$ are nonzero extremal vectors in the same finite module. A sequence of maximal finite simple-root derivatives carries one to the other. Proposition~\ref{prop:maximal-root-derivative} and the chart for $F_0(a)$ show that $f_\theta b_0$ is a monomial in the variables of one seed. We also have $E_0(a)^2b_0=0$. Shifted covariance and Hamiltonian recognition therefore give an adapted chart. \end{proof} \subsection{The affine raising operator in the nonneutral components} In the neutral component, the two affine operators act on the same boundary section. In a nonneutral component their active sections are different: $F_0(a)$ acts on $c=b_{\pi^{-1}(0)}$, whereas $E_0(a)$ acts on $b_0$. We relate these coefficients by finite root strings. \begin{proposition}\label{prop:nonneutral-positive-affine} In types $B_n$ and $D_n$, the translated mutation class contains a seed adapted to $E_0(a)$ in every nonneutral component. \end{proposition} \begin{proof} Evaluation cases follow from Proposition~\ref{prop:evaluation-positive-affine}. Every remaining nonneutral orthogonal case has odd $r$; in type $D_n$, $3\le r\le n-2$. The length-zero permutation interchanges zero and one, so $c=b_1$. Let $V$ be the natural orthogonal module, with basis $v_1,\ldots,v_n,v_{-n},\ldots,v_{-1}$ in type $D_n$, and with the additional vector $v_0$ in type $B_n$. We normalize its symmetric form by $(v_i,v_{-j})=\delta_{ij}$ and, in odd dimension, $(v_0,v_0)=1$. We identify $\g$ with $\bigwedge^2V$ by $ (p\wedge q)v=(q,v)p-(p,v)q$. As a finite $\g$-module, $R_L$ has the multiplicity-free decomposition \[ R_L\simeq\bigwedge^rV\oplus\bigwedge^{r-2}V \oplus\cdots\oplus V. \] At the spin node of type $B_n$, $L=2$ and $\bigwedge^nV=V(2\omega_n)$; otherwise $L=1$. This is the contragredient of the graded decomposition in \cite[Theorem~2.2 and Proposition~2.6]{CM}. Lemma~\ref{lem:finite-boundary-strings} gives $F_i c=0$ and $E_i b_0=0$ for every finite node $i$. Together with their weights $-\omega_1$ and $\omega_1$, these identities make them the lowest and highest vectors of the unique natural summand. This summand is generated from $c$ by finite raising operators. Since translation fixes $c$ and commutes with those operators, it fixes the summand pointwise, including $b_0$. We normalize its identification with $V$ so that $c=v_{-1}$ and $b_0=v_1$; changing either boundary normalization only changes the nonzero scalars below. Here $M=2$, so all currents of degree at least two vanish. The graded decomposition makes the mode-one action on $V$ a map to $\bigwedge^3V$. This map is nonzero: otherwise $V$ would be a current submodule of $R_L$, and its dual would be a quotient of the cyclic KR module, although the cyclic highest weight $L\omega_r$ does not occur in $V$. The space of equivariant maps $\bigwedge^2V\otimes V\to\bigwedge^3V$ is one-dimensional and is spanned by exterior multiplication. Thus $(x\otimes u)v=\xi x\wedge v$, for a fixed $\xi\ne0$. Choose $e_\theta=v_1\wedge v_2$ and $f_\theta=v_{-2}\wedge v_{-1}$. These satisfy $[e_\theta,f_\theta]=h_\theta$. Let $\dot w_{\mathrm{lev}}$ be a Weyl representative in the Levi subgroup with simple roots $\alpha_2,\ldots,\alpha_n$, exchanging $v_2$ with $v_{-2}$ and fixing $v_1,v_{-1}$. Directly from the two actions just described, \[ \dot w_{\mathrm{lev}}(e_\theta c)=-f_\theta b_0,\qquad \dot w_{\mathrm{lev}}((e_\theta\otimes u)c)=(f_\theta\otimes u)b_0. \] It follows that \begin{equation}\label{eq:nonneutral-affine-Levi} \dot w_{\mathrm{lev}}\bigl(F_0(a)c\bigr)=-a^{-2}E_0(a)b_0. \end{equation} Indeed, $F_0(a)c=a^{-1}e_\theta c-a^{-2}(e_\theta\otimes u)c$, whereas $E_0(a)b_0=a f_\theta b_0+(f_\theta\otimes u)b_0$. Both summands of $F_0(a)c$ are highest vectors of natural modules for this Levi subgroup. Their indicated linear combination therefore generates one simple natural Levi module. Its highest vector is a monomial in seed variables by Proposition~\ref{prop:affine-lowering-prefix}. A sequence of maximal finite lowering derivatives carries it to its lowest vector. Proposition~\ref{prop:maximal-root-derivative} and \eqref{eq:nonneutral-affine-Levi} show that $E_0(a)b_0$ is a monomial in the variables of a reachable seed. Finally, $E_0(a)^2b_0$ would have finite weight $\varepsilon_1-2\theta=-\varepsilon_1-2\varepsilon_2$. Every weight of every exterior power occurring in $R_L$ has all Euclidean coordinates in $\{-1,0,1\}$, so this weight is absent. Shifted covariance and Hamiltonian recognition make the seed adapted to $E_0(a)$. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:untwisted-uniform-currents}] Proposition~\ref{prop:translated-finite-roots} gives every finite root chart, and Proposition~\ref{prop:affine-lowering-prefix} gives the $F_0(a)$-chart in the initial translated seed. The positive affine chart follows from Proposition~\ref{prop:positive-affine-from-first-mode} in a neutral component, Proposition~\ref{prop:evaluation-positive-affine} in an evaluation case, and Proposition~\ref{prop:nonneutral-positive-affine} in the remaining classical cases. At the remaining exceptional nodes the common-cluster theorem \cite[Theorem~2.5]{FeiReddening} identifies the translated cluster pattern with the auxiliary spectral pattern considered in the accompanying supplement, up to scalars. The adapted spectral chart for the positive affine generator therefore belongs to the translated mutation class as well. The comparison uses the finite root charts and the $F_0(a)$-chart established above. Hamiltonian recognition gives \eqref{eq:shifted-current-exponents}. Corollary~\ref{cor:framing-symmetrizer} gives the new symmetrizer and the coefficient in the completed $F_r$-chart. \end{proof} \part{Theta bases and crystal structures} \section{Theta bases and current crystals} \label{sec:theta-crystals} We construct finite-type theta crystals for every KR Cartan algebra in Table~\ref{tab:kr-periods}. In the untwisted case, reddening of the translated seeds also gives affine theta crystals. \subsection{Regular theta functions and degree sections} Let $\Sigma$ be a homogeneous seed for $R_r$, with vertex set $V=U\sqcup F$, where $U$ is mutable and $F$ is frozen. Write $N=|V|=\dim R_r$. We use theta functions obtained from principal coefficients by specializing the auxiliary coefficients to one. The original frozen variables are not specialized. A theta function is \emph{finite} when its principal-coefficient lift is a Laurent polynomial, and \emph{regular} when it belongs to $R_r$. We write $\vartheta_g$ for the theta function indexed by $g\in\Z^V$ in the chosen seed. The Cartan degree and finite weight extend to linear maps on this lattice: \[ D(g)=\sum_{v\in V}g_v\deg x_v,\qquad \wt(g)=\sum_{v\in V}g_v\wt(x_v). \] These gradings are compatible with mutation. In our exchange-matrix convention, mutation at a mutable index $k$ transforms its $g$-vector by \[ g'_k=-g_k,\qquad g'_v=g_v+[B_{kv}]_+g_k-B_{kv}[g_k]_+\quad(v\ne k). \] To use the opposite $g$-vector convention, we replace $B$ by $-B$. We normalize $\vartheta_0=1$. A \emph{reddening sequence} is a finite sequence of mutable mutations whose final principal-coefficient $c$-vectors are all nonpositive. \begin{proposition}\label{prop:kr-theta-polytope} Suppose $R_r=\Up(\Sigma)$, the extended exchange matrix has full rank, and the mutable pattern admits a reddening sequence. The regular theta functions form a homogeneous basis of $R_r$. Their $g$-vectors are the integral points of a full-dimensional rational polyhedral cone $\mathscr C_\Sigma\subseteq\mathbb R^V$. The functional $D$ is strictly positive on $\mathscr C_\Sigma\setminus\{0\}$. Consequently $\mathscr P_\Sigma=\mathscr C_\Sigma\cap\{D=1\}$ is a bounded rational polytope of dimension $N-1$, and \[ (R_r)_s=\bigoplus_{g\in s\mathscr P_\Sigma\cap\Z^V}\C\vartheta_g\qquad(s\geq1). \] For $s=0$, the basis consists of $\vartheta_0$. \end{proposition} \begin{proof} Full rank and reddening give the theta basis after inverting the frozen variables \cite[Proposition~1.3]{FeiTheta}. Valuative independence along the frozen divisors implies that a finite theta sum is regular exactly when each term is regular. We therefore obtain a basis of $R_r$ by selecting the regular theta functions. Theta reciprocity expresses each frozen valuation as the minimum of finitely many rational linear forms in $g$. Thus regularity is equivalent to finitely many rational linear inequalities \cite[Corollary~3.2]{FeiTheta}. These inequalities define $\mathscr C_\Sigma$. Each initial coordinate is a regular theta function, so the cone contains all coordinate vectors and is full-dimensional. The exponents in a pointed theta expansion differ by integral combinations of mutable exchange vectors. Compatible gradings annihilate those vectors. Hence $\vartheta_g$ has Cartan degree $D(g)$ and finite weight $\wt(g)$. Since $R_r$ has no negative degrees, $D$ is nonnegative on the integral points of the cone. Rationality then gives nonnegativity on the whole cone. If the face $D=0$ contained a nonzero point, it would contain a nonzero integral point. Its theta function would be a second degree-zero basis element, contrary to $(R_r)_0=\C$. Therefore this face is the origin. The degree-one section is bounded, has dimension $N-1$, and its integral dilates give the stated bases. \end{proof} Replacing the exchange matrix by its negative leaves every exchange relation, and hence the upper cluster algebra, unchanged. Under reddening the two sets of theta functions coincide, although their indexing $g$-vectors generally differ \cite[Proposition~1.5]{FeiTheta}. We use \emph{opposite $g$-vectors} below, meaning the pointed $g$-vectors for the opposite exchange matrix. This convention allows all lowering operators to be treated by the same highest-degree calculation. \subsection{Perfectness and rank-one pairs} For a locally nilpotent operator $X$, define its \emph{nilpotence degree} on a nonzero vector by $\rho_X(v)=\max\{m:X^mv\ne0\}$. A weight basis $\mathcal B$ is \emph{perfect} for $X$ if, for each $b\in\mathcal B$ with $\rho_X(b)>0$, there are $\widetilde Xb\in\mathcal B$ and a nonzero scalar $c_b$ such that $Xb-c_b\widetilde Xb\in\ker X^{\rho_X(b)-1}$, $\rho_X(\widetilde Xb)=\rho_X(b)-1$, and the map $b\mapsto\widetilde Xb$ is injective on its domain. It is \emph{normalized perfect} if $c_b=\rho_X(b)$. We use the following form of theta perfectness \cite[Theorem~7.1]{FeiTheta}. Suppose a seed is adapted to $X=c x^u\partial_b$, where $b$ is frozen, $u_b=0$, and $c\ne0$. For the exchange matrix defining these $g$-vectors, assume that $B_{kb}\ge0$ for every mutable $k$, and that $u_k=0$ whenever $B_{kb}=0$. If the finite theta functions form a basis of the algebra and $X$ preserves it and is locally nilpotent, then its theta basis is perfect for $X$. In the adapted chart the nilpotence degree is the $b$-coordinate of its $g$-vector, and the leading map adds $u-e_b$ to this vector. The operator $c^{-1}X$ is normalized perfect. The multiplier $x^u$ may contain frozen factors. To see the role of the support condition, freeze the mutable coordinates indexed by the nonzero entries of the active column. The highest $b$-degree part of each theta function is a theta function of this smaller mutable pattern. Now $u-e_b$ is supported on frozen coordinates, so multiplication by $x^{u-e_b}$ translates its $g$-vector. Independence of these highest parts gives the unique leading term. This argument applies to the current formulas of Corollary~\ref{cor:kr-origin-currents}. We also recall the rank-one inverse lemma \cite[Lemma~7.4]{FeiTheta}. \begin{lemma}\label{lem:kr-rank-one-perfectness} Let $(E,F,H)$ be an $\mathfrak{sl}_2$-triple acting integrably on a weight module over $\mathbb C$. An $H$-homogeneous normalized perfect basis for $F$ is also normalized perfect for $E$. Its two leading maps are partial inverses. If a basis vector has weight $m$, then its lowering and raising lengths satisfy $\varphi-\varepsilon=m$. \end{lemma} \begin{proof} Decompose the module into finite-dimensional simple $\mathfrak{sl}_2$-modules. On the weight-$m$ space of a summand of highest weight $q$, the raising and lowering lengths are $(q-m)/2$ and $(q+m)/2$. Thus the filtration by highest weights $q$ is the kernel filtration of $F$ on this weight space. Perfectness makes the given basis compatible with that filtration. On each successive quotient, $F$ is an isomorphism between consecutive nonterminal weight spaces. Its matrix in the induced bases has one nonzero entry in each column, so those entries give a bijection. The inverse bijection determines the leading map for $E$. For a vector of lowering length $\varphi$ and raising length $\varepsilon$, the rank-one identity gives $FE=(\varphi+1)\varepsilon$ on its highest nonzero isotypic quotient. The normalized coefficient of $F$ on the preceding weight space is $\varphi+1$, so the leading coefficient of $E$ is $\varepsilon$. Contributions from lower isotypic quotients have raising length at most $\varepsilon-2$. This proves perfectness, inverse arrows, and the string-length identity. Every vector belongs to a finite sum of simple summands, so the argument also applies to an infinite-dimensional integrable module. \end{proof} The geometric pinning may multiply the leading coefficients by nonzero constants. Reciprocal rescaling of $E_i$ and $F_i$ normalizes them without changing the Cartan relations, the Serre relations, or the crystal maps. We apply the lemma to the prescribed current operators, including their frozen monomial factors. \subsection{Finite-type crystals in every current family} Let $\Sigma^\circ$ be an origin seed from Corollary~\ref{cor:kr-origin-cluster}, and use opposite $g$-vectors. Let $\mathcal B^\circ_{r,s}$ denote the $g$-vectors of regular theta functions of Cartan degree $s$. For each finite node $i$, choose a chart adapted to $F_i$. Let $T_i^\circ$ denote tropical mutation to that chart in the opposite pattern, and write $F_i=c_i x^{u_i}\partial_{b(F_i)}$. Define $v_i=u_i-e_{b(F_i)}$. \begin{theorem}\label{thm:origin-theta-crystals} For every current family and node in Table~\ref{tab:kr-periods}, the algebra $R_r$ has the regular theta basis of its Schubert Cox mutation class. This basis is perfect for all finite Chevalley operators and for the positive affine simple operator whenever it is nonzero. For $s\geq0$, its degree-$s$ theta basis defines a finite crystal of type $\g$, with highest-weight decomposition \[ \mathcal B^\circ_{r,s}\simeq\coprod_{\lambda\in P_+}\coprod_{j=1}^{[K_{r,s}:V(\lambda)]}B(\lambda). \] Here $P_+$ is the set of dominant integral weights of $\g$, and $[K_{r,s}:V(\lambda)]$ is the finite $\g$-module multiplicity. For $g\in\mathcal B^\circ_{r,s}$, the string lengths and lowering map are \begin{equation}\label{eq:origin-theta-operators} \varphi_i(g)=(T_i^\circ g)_{b(F_i)},\qquad \varepsilon_i(g)=\varphi_i(g)-\wt(g)(h_i),\qquad \widetilde f_i g=(T_i^\circ)^{-1}(T_i^\circ g+v_i). \end{equation} The last expression applies when $\varphi_i(g)>0$. The raising map is the inverse translation and applies when $\varepsilon_i(g)>0$. Otherwise the corresponding map has value $\mathbf0_{\mathrm{cr}}$. \end{theorem} \begin{proof} The origin seed has full rank and realizes $R_r$ as the upper cluster algebra with non-inverted frozen variables. Its principal matrix is a word matrix, hence admits a maximal green sequence, as does its negative \cite[Lemma~5.1]{FeiSchubert}. Proposition~\ref{prop:kr-theta-polytope} gives the theta basis and its polyhedral parametrization. All gradings in Corollary~\ref{cor:kr-origin-currents} are compatible, so the basis is homogeneous for finite weight and current degree as well as Cartan degree. The same corollary supplies an adapted chart for each $F_i$, with nonpositive active column in the original matrix. It therefore has the required sign for the opposite $g$-vectors. Its exponent formula gives precisely the support condition for theta perfectness. Since these operators preserve $R_r$ and are locally nilpotent, the basis is perfect for every $F_i$. Applying Lemma~\ref{lem:kr-rank-one-perfectness} to each finite rank-one action proves perfectness for $E_i$, inverse arrows, and \eqref{eq:origin-theta-operators}. For a nonzero positive affine operator, its own adapted chart and the original $g$-vectors give perfectness directly. The original and opposite theta functions are the same basis. Every Cartan-degree space is finite dimensional and stable under $\g$. The perfect-basis comparison theorem \cite[Main Theorem~5.55]{BK} identifies its crystal with the disjoint union of the highest-weight crystals of its simple summands. Under \eqref{eq:opposition-convention}, finite duality followed by opposition preserves the highest weight of every simple summand. Its multiplicities are therefore those of $K_{r,s}$, proving the stated decomposition. \end{proof} We call an affine crystal \emph{regular} if its restriction to every finite-type set of colors is a disjoint union of finite-dimensional highest-weight crystals. In finite type, we refer directly to this highest-weight decomposition. The null value $\mathbf0_{\mathrm{cr}}$ is distinct from the lattice origin, which indexes the constant function. \subsection{The shifted affine action} We now work with an untwisted KR Cartan algebra and the translated seed $\Sigma_a$, where $a\ne0$. \begin{theorem}\label{thm:translated-reddening} For every finite simple Lie algebra and every fundamental node, the translated seed $\Sigma_a$, with $a\ne0$, admits a reddening sequence. \end{theorem} \begin{proof} The detailed proof is given in \cite[Theorem~1.1]{FeiReddening}. In an evaluation case, the translated structure is the Schubert structure, whose word matrix admits reddening by \cite[Corollary~4.9]{CaoString}. At the first node of $C_n$, the principal matrix is zero by \eqref{eq:symplectic-exchanges}. For the remaining nodes, the supplement identifies a common cluster in the translated structure and the spectral patterns analyzed in the supplement. Their principal matrices agree up to component signs. The simply laced cases follow from spectral reddening. At a double bond, the local matrix is a weighted two-strip matrix $Q(d_0,d_1)$. Starting from a sequence of 22 mutations for $-Q(2,2)$, the construction extends a reddening word $R$ to $R\,u\,R^{-1}\,v\,R$ whenever its boundary degree is at most two. A symmetric-covering bound and an ordinary-star calculation give this inequality successively at the two ends. They produce $(m,m)$, $(m+1,m)$, and $(m+1,m+1)$ by induction; principal-submatrix heredity then gives every positive pair $(d_0,d_1)$. We apply the covering bound only after constructing a reddening sequence for the smaller matrix. Ordinary-branch completion and the chamber extension in odd orthogonal components give every node of $B$, $C$, and $F_4$. The two $G_2$ cases follow from the explicit maximal green sequences in \cite[Proposition~4.1]{FeiReddening}. Each argument includes $\tau$ when it is mutable. \end{proof} Theorems~\ref{thm:untwisted-translated-cluster} and~\ref{thm:translated-reddening} give full rank and reddening. Proposition~\ref{prop:kr-theta-polytope} therefore supplies the regular theta basis. For seed coordinates we choose representatives of the frozen lines with unit coefficients in the exchange binomials, as permitted by Theorem~\ref{thm:poisson-unfreezing}. In theta functions and frozen monomials, $b_i$ denotes this normalized coordinate. The geometric section $b_i(a)$ spans the same line and has the same current string lengths. We use opposite $g$-vectors and write $\mathscr C_r$, $\mathscr P_r$, and $\mathcal B_{r,s}=s\mathscr P_r\cap\Z^V$ for the corresponding cone, degree-one polytope, and degree-$s$ $g$-vectors; set $\mathcal B_{r,0}=\{0\}$. For each $i\in\widehat I$, choose the chart for $F_i$ in Theorem~\ref{thm:untwisted-uniform-currents}. Write $j_i=\pi^{-1}(i)$, and let $T_i$ be tropical mutation to that chart in the opposite pattern. Write its adapted expression as $F_i=c_i x^{u_i}\partial_{b_{j_i}}$, and set $v_i=u_i-e_{b_{j_i}}$. The weight takes values in the level-zero classical affine weight lattice: its finite part is $\wt(g)$, and $\wt_0(g)=-\wt(g)(h_\theta)$. We also write $\wt_i(g)=\wt(g)(h_i)$ for finite $i$. \begin{theorem}\label{thm:translated-theta-crystals} For every untwisted type, finite node $r$, and $a\ne0$, the regular theta basis of $R_r=\Up(\Sigma_a)$ is perfect for all shifted affine Chevalley generators $E_i(a),F_i(a)$. For every $s\geq0$, the set $\mathcal B_{r,s}$ is a finite regular affine crystal with the character of $(R_r)_s$. Its string functions are \[ \varphi_i(g)=(T_i g)_{b_{j_i}},\qquad \varepsilon_i(g)=\varphi_i(g)-\wt_i(g). \] Its operators are \begin{equation}\label{eq:translated-theta-operators} \widetilde f_i g=T_i^{-1}(T_i g+v_i)\quad(\varphi_i(g)>0),\qquad \widetilde e_i g=T_i^{-1}(T_i g-v_i)\quad(\varepsilon_i(g)>0). \end{equation} Otherwise they have value $\mathbf0_{\mathrm{cr}}$. These formulas are independent of the chosen adapted charts. Every finite-type restriction has the character of the corresponding restriction of $(R_r)_s$. In particular, its finite $\g$-character equals that of the quantum KR module of highest weight $s\omega_r$. \end{theorem} \begin{proof} Proposition~\ref{prop:kr-theta-polytope} and opposite theta invariance give one homogeneous basis on which to compare all currents. Each adapted expression for $F_i$ satisfies the sign and support conditions of theta perfectness. The basis is therefore perfect for all $F_i$, including $F_0(a)$. The current operators form integrable rank-one triples $(E_i,F_i,H_i)$. After reciprocal scalar normalization, Lemma~\ref{lem:kr-rank-one-perfectness} gives perfectness for all $E_i$ and identifies their leading maps as the inverses of those for $F_i$. In an adapted $F_i$-chart, perfectness identifies the lowering length with the active frozen coordinate and the lowering map with translation by $v_i$. The rank-one string identity then gives $\varepsilon_i$ and the inverse translation in \eqref{eq:translated-theta-operators}. This also constructs $\widetilde e_0$ from the chart for $F_0(a)$. The Chevalley--Serre relations are the current relations. All generators preserve Cartan degree. Each degree space is finite dimensional, and its weight decomposition is the one read from the homogeneous theta basis. For any finite-type subset of affine nodes, the restricted integrable module is a direct sum of finite-dimensional simple modules. Applying the perfect-basis comparison theorem \cite[Main Theorem~5.55]{BK} identifies its crystal with the union of their highest-weight crystals. This proves regularity and the character assertion. The opposition convention identifies the finite character with that of $K_{r,s}$, and this is the quantum KR character in every untwisted type by \cite[Corollary~5.1]{KedemPentagon}. The leading basis element in a perfectness expansion is unique. Since the current operators do not depend on the adapted chart, neither do their leading maps or nilpotence degrees. \end{proof} \begin{corollary}\label{cor:kr-theta-joint-kernels} For either theta basis of Theorems~\ref{thm:origin-theta-crystals} and~\ref{thm:translated-theta-crystals}, let $X_1,\ldots,X_m$ be perfect current operators and let $\mathcal B_s$ index the degree-$s$ theta basis. For nonnegative integers $n_1,\ldots,n_m$, \[ (R_r)_s\cap\bigcap_{j=1}^m\ker X_j^{n_j+1}=\bigoplus_{\substack{g\in\mathcal B_s\\ \rho_{X_j}(\vartheta_g)\leq n_j\ (1\leq j\leq m)}}\C\vartheta_g. \] \end{corollary} \begin{proof} Perfectness makes the basis compatible with each kernel filtration; see also \cite[Corollary~7.10]{FeiTheta}. Indeed, iterating the injective leading map sends distinct basis vectors of maximal nilpotence degree to distinct nonzero maximal derivatives. These terms cannot cancel in a linear combination. Each kernel is therefore spanned by the indicated basis vectors, and their intersection has the asserted basis. \end{proof} \section{Cartan degree and minimum level} \label{sec:theta-levels} Throughout this section and Section~\ref{sec:theta-simplicity-limit}, we fix an untwisted type, a finite node $r$, and $a\ne0$. We suppress $a$ from the notation. For the affine theta crystal $\mathcal B_{r,s}$ of Theorem~\ref{thm:translated-theta-crystals}, we write $\varepsilon_i(\vartheta_g)=\varepsilon_i(g)$ and $\varphi_i(\vartheta_g)=\varphi_i(g)$ when describing a theta function by its frozen factorization. We write \[ \varepsilon(g)=\sum_{i\in\widehat I}\varepsilon_i(g)\Lambda_i,\qquad \varphi(g)=\sum_{i\in\widehat I}\varphi_i(g)\Lambda_i. \] The canonical central element is $c_{\mathrm{aff}}=\sum_i k_i h_i$. We define the \emph{level} of an element by $\ell(g)=\langle c_{\mathrm{aff}},\varepsilon(g)\rangle =\sum_i k_i\varepsilon_i(g)$. Since the current action has level zero, this also equals $\sum_i k_i\varphi_i(g)$. \subsection{Primitive theta functions} We use the opposite seed. The charts adapted to the lowering operators are \emph{optimizing seeds} for their active frozen variables: the mutable entries of each active column are nonnegative. Let $U$ and $F=\widehat I$ be the mutable and frozen index sets. To apply the support-function formulas of~\cite{FeiTheta}, we write $\mathsf B$ for the square exchange matrix and choose positive integers $\sigma_v$ proportional to $\delta_v^{-1}$. Then $\mathsf D=\diag(\sigma_v)$ is a right symmetrizer: $\mathsf B\mathsf D$ is skew-symmetric. In particular, $\sigma_jt_j$ is independent of the frozen index $j$. The entries with two frozen indices will not be used. Let $\mathbf d_U$ be the column of mutable Cartan degrees and set $a_j=\mathsf B_{j,U}$. For $p\in\Z^U$, there is a unique regular theta function with mutable $g$-vector $p$ whose order along every frozen divisor is zero. We denote it by $\chi_p$ and call it the \emph{primitive regular theta function} at $p$. Frozen factorization gives \[ R_r=\bigoplus_{p\in\Z^U}\C[b_j:j\in F]\chi_p. \] Here and below a frozen monomial has nonnegative exponents. This is the frozen factorization of \cite[Proposition~5.1]{FeiTheta}. We write $d(p)=\deg\chi_p$. Since $(R_r)_0=\C$, we have $d(p)\ge0$, with equality only for $p=0$. We recall the support functions needed for the degree calculation. Let $F_z$ be the \emph{principal-coefficient $F$-polynomial} of mutable theta $g$-vector $-z$, and let $f_z$ be its componentwise largest exponent \cite[Propositions~1.3 and~1.7]{FeiTheta}. For $y,z\in\Z^U$, regarded as row vectors, define \[ \mathsf E(y,z)=\max_{\nu\in\supp F_z}(-y\mathsf D_U\nu^{\mathsf T}),\qquad \mathsf H(y,z)=y\mathsf D_Uf_z^{\mathsf T}+\mathsf E(y,z). \] We use the inverse orientation of the mutable tropical Donaldson--Thomas transformation. It is the bijection $\mathsf T$ characterized by \[ \mathsf T^{-1}(z)=-z-f_z\mathsf B_{U,U}. \] Theta reciprocity gives $\mathsf E(y,z)=\mathsf H(z,\mathsf T y)$ \cite[Theorem~4.3]{FeiTheta}. On mutable $g$-vectors, we use $\mathsf S(p)=-\mathsf T(-p)$, an integral piecewise-linear bijection fixing the origin. The frozen coordinates of $\chi_p$ are $\beta_j(p)=\mathsf E(a_j,-p)/\sigma_j$. Thus \begin{equation}\label{eq:theta-primitive-degree} d(p)=p\mathbf d_U+ L\sum_{j\in F}\frac{k_j}{\sigma_j}\mathsf E(a_j,-p). \end{equation} Although the support functions use the symmetrizer, the quotients $\beta_j(p)$ are integers: they are precisely the powers required to cancel the frozen valuations. \begin{lemma}\label{lem:theta-string-support} For $p\in\Z^U$ and $\mathbf q\in\Z_{\ge0}^F$, \[ \varphi_{\pi(j)}(b^{\mathbf q}\chi_p)=q_j+\frac{\mathsf E(a_j,-p)+\mathsf E(-p,a_j)}{\sigma_j}\qquad(j\in F). \] \end{lemma} \begin{proof} Let $M_j$ be the canonical mutable multiplier in an optimizing seed for $j$ in the opposite pattern. Its mutable $g$-vector in the reference seed is $-a_j$ \cite[Lemma~2.1]{FeiTheta}. The symmetric theta $F$-invariant of $M_j$ and $\chi_p$ is $\mathsf E(a_j,-p)+\mathsf E(-p,a_j)$ \cite[Proposition~9.3]{FeiTheta}. We evaluate it in the optimizing seed, where $F_{M_j}=1$. Let $\gamma$ be the $g$-vector of $\chi_p$ in that seed and let $m\ge0$ be the mutable exponent vector of $M_j$. A term indexed by $\nu$ in the pointed expansion of $\chi_p$ has $b_j$-exponent $\gamma_j-\sigma_j^{-1}\nu\mathsf D_Um^{\mathsf T}$. Its maximum is $\gamma_j$, from the constant term, and its minimum is zero because the frozen valuation of $\chi_p$ vanishes. The invariant is therefore $\sigma_j\gamma_j$, or $\sigma_j$ times the degree in $b_j$. The multiplier of $F_{\pi(j)}$ is a scalar times $M_j$ times a frozen monomial independent of $b_j$. Its nilpotence degree is consequently the degree in $b_j$. Multiplication by $b^{\mathbf q}$ adds $q_j$ to that degree, proving the formula. \end{proof} \begin{theorem}\label{thm:theta-degree-level} For every regular theta function $\vartheta_g=b^{\mathbf q}\chi_p$, \begin{equation}\label{eq:theta-degree-level} L\ell(g)=\deg\vartheta_g+d(\mathsf S(p)). \end{equation} Consequently $L\ell(g)\ge\deg\vartheta_g$, with equality if and only if $\vartheta_g$ is a frozen monomial. \end{theorem} \begin{proof} Homogeneity of the exchanges and skew-symmetrizability give \begin{equation}\label{eq:theta-degree-compatibility} \mathsf B_{U,U}\mathbf d_U =L\mathsf D_U\sum_{j\in F}\frac{k_j}{\sigma_j}a_j^{\mathsf T}. \end{equation} Indeed, the frozen degrees are $Lk_j$, while $\mathsf B_{uj}=-(\sigma_u/\sigma_j)a_j(u)$. Let $z=-p$ and $t=\mathsf T z$, so that $z=-t-f_t\mathsf B_{U,U}$. The permutation $\pi$ preserves the comarks. Hence Lemma~\ref{lem:theta-string-support} and \eqref{eq:theta-primitive-degree} imply \[ L\ell(g)-\deg\vartheta_g =L\sum_{j\in F}\frac{k_j}{\sigma_j}\mathsf E(z,a_j) +z\mathbf d_U. \] Apply the duality identity $\mathsf E(z,a_j)=\mathsf H(a_j,t)$, then \eqref{eq:theta-degree-compatibility}. The right side becomes \begin{align*} &L\sum_{j\in F}\frac{k_j}{\sigma_j} \mathsf E(a_j,t)+f_t\mathsf B_{U,U}\mathbf d_U +z\mathbf d_U\\ &\hspace{2em}= L\sum_{j\in F}\frac{k_j}{\sigma_j}\mathsf E(a_j,t) -t\mathbf d_U =d(-t)=d(\mathsf S(p)). \end{align*} This proves \eqref{eq:theta-degree-level}. The grading is positive, so the last degree is nonnegative and vanishes only when $\mathsf S(p)=0$. Since $\mathsf S$ is bijective and fixes zero, this is equivalent to $p=0$. The primitive theta function at zero is the constant function. \end{proof} \subsection{Minimal elements and simultaneous kernels} For $s\ge0$, let $Q_s=\{p\in\Z^U:d(p)\le s,\ d(p)\equiv s\pmod L\}$. This set is nonempty because $(R_r)_s\ne0$. It is finite because its primitive theta functions belong to $\bigoplus_{j=0}^s(R_r)_j$. \begin{corollary}\label{cor:theta-minima} The minimum level in $\mathcal B_{r,s}$ is \[ \min_{g\in\mathcal B_{r,s}}\ell(g)=\frac{s+\min_{p\in Q_s}d(\mathsf S(p))}{L}\ \ge\left\lceil\frac{s}{L}\right\rceil. \] Its minimal elements are exactly the functions $b^{\mathbf q}\chi_p$ for which $p$ realizes the indicated minimum and $L\sum_jk_jq_j=s-d(p)$. If $s=L\ell$, the minimum level is $\ell$, and its minimal elements are precisely the frozen monomials $b^{\mathbf q}$ with $\sum_jk_jq_j=\ell$. On this set, both $\varepsilon$ and $\varphi$ are bijections onto the dominant classical affine weights of level $\ell$. \end{corollary} \begin{proof} Since $k_0=1$, every nonnegative multiple of $L$ is the degree of a frozen monomial. Thus $p\in Q_s$ is exactly the condition that some $b^{\mathbf q}\chi_p$ have degree $s$. Formula~\eqref{eq:theta-degree-level} gives the minimum and its stated parametrization. When $s=L\ell$, the frozen monomial $b_0^\ell$ attains level $\ell$. Equality in Theorem~\ref{thm:theta-degree-level} characterizes all minimal elements. The boundary strings give \[ \varepsilon_i(b^{\mathbf q})=q_i,\qquad \varphi_i(b^{\mathbf q})=q_{\pi^{-1}(i)}. \] The first equality also follows from the second and the weight $\wt(b_j)=\Lambda_{\pi(j)}-\Lambda_j$. The comark-preserving permutation $\pi$ now gives both bijections. \end{proof} When $L\nmid s$, frozen monomials do not occur in degree $s$; the first assertion describes the minimal elements in these degrees. \begin{corollary}\label{cor:theta-joint-kernels} For $\mathbf q\in\Z_{\ge0}^{\widehat I}$, the space $(R_r)_s\cap\bigcap_{i\in\widehat I}\ker E_i^{q_i+1}$ has the theta basis indexed by those $g\in\mathcal B_{r,s}$ with $\varepsilon_i(g)\le q_i$ for every $i$. It is zero if $s>L\sum_i k_i q_i$, and equals $\C b^{\mathbf q}$ if $s=L\sum_i k_i q_i$. For the lowering operators, replace $\varepsilon_i$ by $\varphi_i$; at equality the spanning monomial is $\prod_j b_j^{q_{\pi(j)}}$. \end{corollary} \begin{proof} Corollary~\ref{cor:kr-theta-joint-kernels} gives the asserted basis. For an index in the intersection, $s\le L\sum_i k_i\varepsilon_i(g)\le L\sum_i k_i q_i$. This proves vanishing above the indicated degree. At equality, Theorem~\ref{thm:theta-degree-level} makes the theta function a frozen monomial, and positivity of the comarks gives $\varepsilon_i(g)=q_i$ for every $i$. There is exactly one such monomial. The lowering assertion uses the same argument and the formula for $\varphi$ in Corollary~\ref{cor:theta-minima}. \end{proof} In particular, if $f\in R_r$ is nonzero and homogeneous, then \[ \deg f\le L\sum_i k_i\max\{m:E_i^mf\ne0\}. \] Equality holds precisely for scalar multiples of frozen monomials. Indeed, apply Corollary~\ref{cor:theta-joint-kernels} with the actual nilpotence orders of $f$. \section{Simplicity and the common limit} \label{sec:theta-simplicity-limit} We prove simplicity using the simple head of the current module. We then apply known uniqueness theorems to identify the classical, minuscule, and adjoint cases with quantum KR crystals, and construct a common limit. \subsection{An essential-socle criterion} Let $W_0$ be the finite Weyl group and $Q_+=\sum_{i=1}^n\Z_{\ge0}\alpha_i$. We use the term \emph{simple crystal} in the sense of Akasaka--Kashiwara~\cite[Definition~1.7]{AK}. \begin{lemma}\label{lem:theta-simple-socle} Let $V$ be a finite-dimensional loop-algebra module of level zero with a weight basis perfect for all affine Chevalley generators and with regular crystal. Suppose that its weights belong to $\lambda_0-Q_+$, where $\lambda_0$ is dominant, that $\dim V_{\lambda_0}=1$, and that every nonzero submodule contains $V_{\lambda_0}$. Then its crystal is simple. \end{lemma} \begin{proof} Suppose $b$ is the highest element of an $i$-string of length $m$, and let $v_b$ be its basis vector. Perfectness implies that $F_i^mv_b$ is a nonzero scalar multiple of the basis vector at the bottom of the string: every lower term is killed by the remaining lowering operators. Rank-one representation theory therefore shows that the Tits operator $\dot s_i=\exp(E_i)\exp(-F_i)\exp(E_i)$ sends $v_b$ exactly to a scalar multiple of the basis vector at $s_i b$. The same argument applies at the bottom of a string. We can therefore lift the crystal Weyl group action on an extremal element to the corresponding basis vector, up to nonzero scalar. Let $b$ be extremal. After a finite Weyl transformation, we may suppose that $\lambda=\wt(b)$ is dominant for $\g$. For a positive finite root $\alpha$ and $m\in\Z$, choose an affine Weyl element $w$ and a simple affine root $\alpha_i$ such that $\alpha+m\delta_{\mathrm{aff}}=w\alpha_i$. The coroot pairing with $\lambda$ is nonnegative because the module has level zero. Extremality gives $\widetilde e_i(w^{-1}b)=\mathbf0_{\mathrm{cr}}$. Perfectness and the preceding compatibility with the Weyl group give $ (e_\alpha\otimes t^m)v_b=0$. The loop triangular decomposition implies that every finite weight of the submodule generated by $v_b$ lies in $\lambda-Q_+$. This submodule contains $V_{\lambda_0}$, so $\lambda_0\le\lambda$. The assumed weight bound gives the reverse inequality. Hence every extremal weight belongs to $W_0\lambda_0$. The multiplicity-one hypothesis and Weyl invariance give a unique element at each weight in this orbit. The same weight bound, together with finite Weyl invariance, places all weights in the convex hull of $W_0\lambda_0$. These are the remaining conditions for simplicity. \end{proof} \begin{theorem}\label{thm:theta-kr-simple} For every untwisted type and finite node $r$, the crystal $\mathcal B_{r,s}$ is simple for every $s>0$. Every finite tensor product of such crystals within one affine type is connected. The same assertions hold for their duals. For $s=L\ell$, they satisfy conditions \textup{(ii)--(v)} of a level-$\ell$ perfect crystal in \cite[Definition~2.4]{ST}. \end{theorem} \begin{proof} The ideal $\g\otimes u\C[u]$ strictly raises the bounded current grading of $K_{r,s}$, whereas every current operator preserves or raises it. Its image therefore generates a nilpotent ideal in the associative algebra acting on $K_{r,s}$. Thus every simple quotient factors through \[ K_{r,s}/(\g\otimes u\C[u])K_{r,s}\simeq V(s\omega_r). \] The equality follows from the finite highest-weight relations and the nonzero evaluation quotient. Thus $K_{r,s}$ has simple head. Its dual has an essential simple socle, since every nonzero finite-dimensional submodule contains a simple submodule. With the opposition convention \eqref{eq:opposition-convention}, this socle has highest weight $s\omega_r$. All weights of $(R_r)_s$ lie in $s\omega_r-Q_+$, and the top weight has multiplicity one. The substitution $t=a+u$ maps the loop algebra onto the finite jet algebra acting on $K_{r,s}$. Thus its loop submodules are precisely its current submodules. Theorem~\ref{thm:translated-theta-crystals} and Lemma~\ref{lem:theta-simple-socle} now prove simplicity. Simple crystals are connected and closed under tensor product \cite[Lemmas~1.9--1.10]{AK}; dualization preserves simplicity. Finally, the highest-weight bound and Corollary~\ref{cor:theta-minima} give the stated perfectness conditions. The following corollaries supply the quantum realization required by condition \textup{(i)} in the stated cases. \end{proof} \begin{corollary}\label{cor:theta-classical-quantum} For every node $r$ of classical type and every $s>0$, the crystal $\mathcal B_{r,s}$ is isomorphic to the quantum Kirillov--Reshetikhin crystal $B^{r,s}$. It is perfect if and only if $L\mid s$, and then has level $s/L$. \end{corollary} \begin{proof} By Theorem~\ref{thm:translated-theta-crystals}, $\mathcal B_{r,s}$ and $B^{r,s}$ have the same finite character. Their highest-weight decompositions therefore agree on the finite colors. Schilling--Tingley's uniqueness lemma gives the affine isomorphism \cite[Theorem~2.5 and Lemma~2.6]{ST}. In type $A_1$, the finite restriction is one $\mathfrak{sl}_2$-string, and the affine string reverses it, giving the same conclusion directly. Perfectness and its level are the theorem of Fourier--Okado--Schilling~\cite[Theorem~1.2]{FOSPerfect}. \end{proof} \begin{corollary}\label{cor:theta-exceptional-quantum} For every $s>0$, the identification $\mathcal B_{r,s}\simeq B^{r,s}$ holds in the following cases. \[ \begin{array}{c|c} \text{finite type}&r\\ \hline E_6&1,2,6\\ E_7&1,7\\ E_8&8\\ F_4&1\\ G_2&2. \end{array} \] These crystals are perfect of level $s$. \end{corollary} \begin{proof} At each listed node, $t_r=1$. The current KR module is therefore a Demazure module by \cite[Theorem~4]{FL}. Its finite character agrees with that of the quantum KR module by \cite[Theorem~2 and Corollary~1]{FLTensor}. The quantum KR module admits a \emph{crystal pseudobase}: the minuscule cases follow from \cite[Proposition~3.4.4]{KKM}, and the adjoint cases from \cite[Proposition~3.4.5]{KKM}. Its associated crystal $B^{r,s}$, obtained by identifying opposite signs, therefore has the same finite character as $\mathcal B_{r,s}$. Both crystals have level-zero weights. Their finite characters therefore determine their full weight characters modulo $\delta_{\mathrm{aff}}$, since the central relation determines the zeroth coroot pairing. Regularity then determines their restrictions to every proper set of affine colors: the character of a finite-type restriction determines its highest-weight decomposition. For the minuscule nodes $E_6,1$, $E_6,6$, and $E_7,7$, the finite restriction is $B(s\omega_r)$. The uniqueness theorems of Jones--Schilling \cite[Theorem~3.15 and Corollary~3.16]{JS} and Biswal--Scrimshaw~\cite[Theorem~3.5]{BS7} now identify the affine crystals. The additional finite restrictions in those theorems agree by the preceding character argument. At the adjoint nodes, $\omega_r=\theta$, and the finite restriction is $\bigsqcup_{k=0}^s B(k\theta)$. The affine node has a unique neighbor, joined to it by a single bond. The uniqueness theorem \cite[Theorem~6.1]{KMOY} therefore applies. The proof requires only the finite highest-weight decompositions and the crystal Weyl group action, so it applies to our regular crystals; this formulation is also used in \cite[Corollary~3.23]{JS}. This proves the asserted identifications. Since $L=1$, Theorem~\ref{thm:theta-kr-simple} and Corollary~\ref{cor:theta-minima} give the remaining perfectness conditions at level $s$. \end{proof} \subsection{A limit on the degree-zero lattice} Let $\mathsf M=\Z^V$ be the full $g$-vector lattice of the opposite reference seed. Its compatible gradings are $D:\mathsf M\to\Z$ and $\wt:\mathsf M\to P_{\mathrm{cl}}$. We define the limit on the degree-zero lattice $\mathcal B_{r,\infty}=\ker D$. For each $i\in\widehat I$, choose a chart adapted to $F_i$, with active frozen index $j=\pi^{-1}(i)$. Let $T_i$ be the tropical change to that chart, using the opposite pattern, and let $v_i=u_i-e_j$, where $x^{u_i}$ is the full monomial multiplier of $F_i$. Define maps on the whole lattice by \[ \widetilde f_i^{\infty}(g)=T_i^{-1}(T_i(g)+v_i),\qquad \widetilde e_i^{\infty}(g)=T_i^{-1}(T_i(g)-v_i). \] Their string functions are \[ \varphi_i^{\infty}(g)=(T_i(g))_j,\qquad \varepsilon_i^{\infty}(g)=\varphi_i^{\infty}(g)-\langle h_i,\wt(g)\rangle. \] These functions may be negative. Neither operator has a null value on $\mathcal B_{r,\infty}$. For $\mathbf q\in\Z_{\ge0}^{\widehat I}$, write $\Lambda(\mathbf q)=\sum_i q_i\Lambda_i$ and $(\pi\mathbf q)_i=q_{\pi^{-1}(i)}$. We use the tensor convention in which $\widetilde e_i$ acts on the left factor when $\varphi_i(b_1)\ge\varepsilon_i(b_2)$, and $\widetilde f_i$ does so when the inequality is strict. We identify a frozen exponent vector with its extension by zero on the mutable indices in $\mathsf M$. Let $T_\lambda=\{t_\lambda\}$ be the one-element crystal of weight $\lambda$, with string functions $-\infty$. \begin{proposition}\label{prop:theta-common-limit} The preceding formulas make $\mathcal B_{r,\infty}$ an abstract crystal, independently of the adapted charts. For $\sum_i k_i q_i=\ell>0$, the map \begin{equation}\label{eq:theta-limit-embedding} T_{\Lambda(\mathbf q)}\otimes\mathcal B_{r,L\ell} \otimes T_{-\Lambda(\pi\mathbf q)} \longrightarrow\mathcal B_{r,\infty},\qquad t_{\Lambda(\mathbf q)}\otimes g\otimes t_{-\Lambda(\pi\mathbf q)} \longmapsto g-\mathbf q \end{equation} preserves weights, string functions, and every arrow in its source. Each map sends the element indexed by $b^{\mathbf q}$ to zero, and their images cover $\mathcal B_{r,\infty}$. At the nodes of Corollaries~\ref{cor:theta-classical-quantum} and~\ref{cor:theta-exceptional-quantum}, this is the limit of a coherent family of perfect KR crystals. \end{proposition} \begin{proof} Tropical mutation commutes with every frozen translation. Since $v_i$ has Cartan degree zero and classical affine weight $-\alpha_i$, the two operators are inverse bijections of $\ker D$ with the required weight changes. The active coordinate of $v_i$ is $-1$, so $\varphi_i^{\infty}$ decreases by one under $\widetilde f_i^{\infty}$. The definition of $\varepsilon_i^{\infty}$ gives the remaining crystal axioms. For any integral frozen vector $\mathbf q$, we have $\wt(\mathbf q)=\Lambda(\pi\mathbf q)-\Lambda(\mathbf q)$. Together with frozen translation, this gives \begin{align*} \varepsilon_i^{\infty}(g+\mathbf q)&=\varepsilon_i^{\infty}(g)+q_i,\\ \varphi_i^{\infty}(g+\mathbf q)&=\varphi_i^{\infty}(g)+q_{\pi^{-1}(i)}. \end{align*} A sufficiently large nonnegative $\mathbf q$ makes any fixed finite collection of theta functions regular and their relevant string lengths positive. On the corresponding translated $g$-vectors the formulas agree by Theorem~\ref{thm:translated-theta-crystals}. Subtracting $\mathbf q$ then compares any two adapted charts. This proves independence of the charts on the whole lattice. The origin has weight zero and both string functions vanish there. For $g$ in the source of \eqref{eq:theta-limit-embedding}, we have $D(g)=D(\mathbf q)=L\ell$. The preceding weight and string identities agree exactly with the tensor-product formulas, and frozen translation commutes with every source arrow. The map is injective, so it is a crystal embedding in the sense of \cite[Definition~3.2]{KKHead}. It sends the element indexed by $b^{\mathbf q}$ to zero. Arrows of the limit may leave a finite image. Finally, given $g\in\ker D$, choose a sufficiently large nonnegative frozen vector $\mathbf q$ so that $g+\mathbf q$ is regular and $\sum_i k_iq_i=\ell>0$. Then $g+\mathbf q\in\mathcal B_{r,L\ell}$, and its image is $g$. Thus the embeddings cover the degree-zero lattice and satisfy the limit axioms of \cite[Definition~5.2]{KKHead}. In the cases with quantum realization, the finite crystals are perfect and the family is coherent. \end{proof} \section{Two examples}\label{sec:examples} The first example makes the translated exchange relations and the crystal-level formula explicit. The second illustrates the primitive Cartan grading in the even twisted family. The end-node evaluation case in type~$A$ is the polynomial coordinate ring of projective space and needs no further calculation. \subsection{The first node of type \texorpdfstring{$C_n$}{Cn}} Let $V$ be the natural symplectic module, with $n\geq2$. The first-node KR Cartan algebra is \begin{equation}\label{eq:symplectic-polynomial-ring} R=\C[x_1,\ldots,x_n,y_1,\ldots,y_n,z],\qquad \deg x_i=\deg y_i=1,\quad\deg z=2. \end{equation} The finite Chevalley operators are \[ E_i=x_i\partial_{x_{i+1}}-y_{i+1}\partial_{y_i},\qquad F_i=x_{i+1}\partial_{x_i}-y_i\partial_{y_{i+1}}\quad(i0$, and write $v_i=(-1)^{i-1}y_i$. The translated seed has mutable coordinates $x_1,\ldots,x_n$, frozen coordinates $c_0,\ldots,c_n$, and exchanges \begin{equation}\label{eq:symplectic-exchanges} x_iv_i=c_{i-1}+c_i\qquad(1\leq i\leq n). \end{equation} Its principal exchange matrix is zero. The $2^n$ seeds arise by choosing $x_i$ or $v_i$ independently. Their mixed Laurent rings are the corresponding localizations of $R$. These charts meet every height-one prime: no such prime contains both $x_i$ and $y_i$ for any $i$. Normality therefore gives $R=\Up(\Sigma_a)$ directly. The shifted affine operators are \[ E_0(a)=ay_1\partial_{x_1}+y_1^2\partial_z,\qquad F_0(a)=a^{-1}x_1\partial_{y_1}+a^{-2}x_1^2\partial_z. \] Their values on the active frozen coordinates, together with the finite-root formulas, are \[ \begin{aligned} E_0(a)b_0&=y_1^2,&F_0(a)b_0&=a^{-2}x_1^2,\\ E_i b_i&=x_iy_{i+1},&F_i b_i&=-x_{i+1}y_i\quad(i0}\bigoplus_{j=0}^{m_\alpha-1} \C f_\alpha u^j, \qquad N_\lambda=\exp(\mathfrak n_\lambda). \] The sum for a root with $m_\alpha=0$ is empty. The inequality $j+k0}t_\alpha^{-1}e_\alpha\wedge f_\alpha$. After interchanging the Borels, this gives the finite term $-E_i\wedge F_i/(2t_i)$ used in the main text. We construct the tensors on the loop group and descend them to the framed-bundle quotient. Let $\mathfrak d=\mathfrak g\oplus\h$, with invariant form \[ ((x,h),(y,k))_{\mathfrak d}=(x,y)_{\mathfrak g}-(h,k)_{\h}. \] The two subalgebras \[ \mathfrak l=\{(x_-+h,h):x_-\in\mathfrak n_-,\ h\in\h\}, \qquad \mathfrak l'=\{(x_++h,-h):x_+\in\mathfrak n_+,\ h\in\h\} \] are complementary and Lagrangian. Thus $(\mathfrak d,\mathfrak l,\mathfrak l')$ is a \emph{Manin triple}. The subgroup integrating $\mathfrak l$ is naturally isomorphic to $B_-$. For dual bases $a_i\in\mathfrak l$ and $a^i\in\mathfrak l'$, let $r_{\mathfrak d}=\frac12\sum_i a_i\wedge a^i$. The standard double construction \cite[Sections~1.3B, 1.4B, and~2.2]{ChariPressley} gives the corresponding multiplicative Poisson tensor, in right-trivialized tangent coordinates, \[ \rho(b)=(\Ad_b\otimes\Ad_b)r_{\mathfrak d}-r_{\mathfrak d}. \] It takes values in $\bigwedge^2\mathfrak l$. Since $r_{\mathfrak d}$ is torus invariant, $\rho$ vanishes on the torus. For a nonzero affine polynomial $p(u)=cu+d$, consider the loop Manin triple \[ (\mathfrak d((u^{-1})),\mathfrak l((u^{-1})), \mathfrak l'((u^{-1}))), \qquad \langle x,y\rangle_p =-\Res_{u=\infty}(x(u),y(u))_{\mathfrak d} \frac{du}{p(u)}. \] In this pairing, $\mathfrak l'((u^{-1}))$ is the continuous dual Lie algebra of $\mathfrak l((u^{-1}))$. The annihilators in this dual Lie algebra of the two gauge algebras are \[ \mathfrak l[u]^{\mathrm{ann}}=p(u)\mathfrak l'[u], \qquad \bigl(u^{-1}\mathfrak l[[u^{-1}]]\bigr)^{\mathrm{ann}} =p(u)u^{-1}\mathfrak l'[[u^{-1}]]. \] Both are Lie subalgebras: the first because $p$ is polynomial, and the second because $\deg p\le1$. The two gauge groups are consequently \emph{coisotropic}, so functions invariant under both gauge actions form a Poisson algebra. We obtain a Poisson tensor $\Pi_p$ on the framed-bundle quotient \[ B_{-,1}[[u^{-1}]]\backslash B_-((u^{-1}))/B_-[u]. \] The degree of the torus bundle is locally constant in families. Thus $N_\lambda$ is an open-and-closed component, and $\Pi_p$ restricts to it. To compute the quotient tensor, let $Q$ denote polynomial part and let $P_\lambda:\mathfrak l[u]\to\mathfrak n_\lambda$ be the projection in \[ \mathfrak l[u]=\mathfrak n_\lambda \oplus\Ad_{u^{-\lambda}}\mathfrak l[u]. \] We regard $\mathfrak n_\lambda$ as a subspace of the first summand of $\mathfrak d$. Extend $P_\lambda$ by zero on $\mathfrak l'[u]$. At $b=n(u)u^{-\lambda}$, the differential of based normalization, with left-trivialized target tangent space, is \begin{equation}\label{eq:normalization-differential} D_n=P_\lambda\Ad_{n^{-1}}Q. \end{equation} Indeed, differentiating a factorization $b=b_\infty n u^{-\lambda}b_0$ at normalized factors gives \[ \dot b b^{-1} =\xi_\infty+\dot n n^{-1} +\Ad_{n u^{-\lambda}}\xi_0, \qquad \xi_\infty\in u^{-1}\mathfrak l[[u^{-1}]],\quad \xi_0\in\mathfrak l[u]. \] Polynomial part removes the first term. Conjugation by $n^{-1}$, followed by $P_\lambda$, removes the last term and gives $n^{-1}\dot n$. This proves \eqref{eq:normalization-differential}. The current Poisson tensor has kernel $p(u)\rho(b(u))\delta(u,v)$, where $\delta(u,v)=\sum_{k\in\Z}u^kv^{-k-1}$. For every polynomial tensor $H(u)$, polynomial part in both variables gives the finite identity \begin{equation}\label{eq:polynomial-delta-projection} (Q\otimes Q)\bigl(H(u)\delta(u,v)\bigr) =\frac{H(u)-H(v)}{u-v}. \end{equation} To check it, take $H(u)=u^j H_0$. The surviving terms are $\sum_{k=0}^{j-1}u^{j-1-k}v^k H_0$; the sum is empty when $j=0$. Since $\rho(u^{-\lambda})=0$, multiplicativity gives $\rho(nu^{-\lambda})=\rho(n)$ in right-trivialized coordinates. Equations \eqref{eq:normalization-differential} and \eqref{eq:polynomial-delta-projection} therefore give \begin{equation}\label{eq:framed-divided-difference} \Pi_p(n) =(P_\lambda\otimes P_\lambda) (\Ad_{n(u)^{-1}}\otimes \Ad_{n(v)^{-1}}) \frac{p(u)\rho(n(u))-p(v)\rho(n(v))}{u-v}. \end{equation} This formula is polynomial in the coordinates of $n$, so $\Pi_p$ is regular on the cell. Let $\Omega_{\mathfrak d}$ be the Casimir tensor of the quadratic double. We have \[ r_{\mathfrak d} =\frac12\Omega_{\mathfrak d}-\sum_i a^i\otimes a_i =\sum_i a_i\otimes a^i-\frac12\Omega_{\mathfrak d}. \] Substitute the formula for $\rho$ into \eqref{eq:framed-divided-difference}. In its term containing $p(u)$, use the first expression for $r_{\mathfrak d}$; after conjugation, the first tensor factor $a^i$ is annihilated by $P_\lambda$. In the term containing $-p(v)$, use the second expression; its second factor $a^i$ is annihilated. Invariance of the Casimir tensor then gives \[ \Pi_p(n)=(P_\lambda\otimes P_\lambda) (\Ad_{n(u)^{-1}}\otimes \Ad_{n(v)^{-1}}) \left( \frac{p(u)+p(v)}{2(u-v)}\Omega_{\mathfrak d} -c r_{\mathfrak d} \right). \] The auxiliary Cartan summand of $\mathfrak d$ is fixed by $n$ and is annihilated by $P_\lambda$. Thus $\Omega_{\mathfrak d}$ reduces to $\Omega_{\mathfrak g}$, while $-r_{\mathfrak d}$ reduces to $\frac12\sum_{\alpha>0}t_\alpha^{-1}e_\alpha\wedge f_\alpha$. Consequently the kernel on $\mathfrak g$ is exactly \begin{equation}\label{eq:general-affine-kernel} \mathcal R_p(u,v)=\frac{p(u)+p(v)}{2(u-v)}\Omega_{\mathfrak g} +\frac c2\sum_{\alpha>0}t_\alpha^{-1}e_\alpha\wedge f_\alpha. \end{equation} In the Casimir term, after conjugation we first project to the two negative root spaces. Casimir invariance makes the numerator vanish at $u=v$, so division gives a polynomial before we extract the prescribed coefficients. For $p=1$, \eqref{eq:general-affine-kernel} is the rational kernel. For $p=u$, it is the trigonometric kernel of \cite[Equation~(4.11)]{FKR} in the negative-Borel model, after rescaling the invariant form from their short-root normalization to the long-root normalization used here. Applying $P_\lambda\otimes P_\lambda$ after conjugation is the homogeneous-space formula for these tensors. The term at the coweight point vanishes because $P_\lambda$ kills the positive root spaces and the Cartan subalgebra. Hence $\Pi_1=\Pi_{\mathrm{rat}}$ and $\Pi_u=\Pi_{\mathrm{tr}}$. Moreover $\Pi_{cu+d}=c\Pi_{\mathrm{tr}}+d\Pi_{\mathrm{rat}}$. The quotient construction proves Jacobi for every nonzero $p$, so the two tensors are compatible. Finally, source pullback preserves both gauge groups and transports the current tensor by the corresponding change of vector field $p(u)\partial_u$. In the action on functions induced by $u\mapsto u-a$, it follows that $\Psi_a\Pi_p=\Pi_{p(u+a)}$. The stated identities follow by taking $p=1$ and $p=u$. The same argument gives covariance under affine changes of coordinate. \end{proof} \section*{Declaration of generative AI and AI-assisted technologies} ChatGPT was used for mathematical exploration, proof development, exact computations, and revision of the exposition. Its contributions include the uniform reddening argument and the accompanying verification programs. \begin{thebibliography}{99} \bibitem[AB]{AbedinBurban} R. Abedin and I. Burban, \emph{Algebraic geometry of Lie bialgebras defined by solutions of the classical Yang--Baxter equation}, Comm. Math. Phys. \textbf{387} (2021), 1051--1109. \href{https://doi.org/10.1007/s00220-021-04188-7}{doi:10.1007/s00220-021-04188-7}. \bibitem[AK]{AK} T. Akasaka and M. Kashiwara, \emph{Finite-dimensional representations of quantum affine algebras}, Publ. Res. Inst. Math. Sci. \textbf{33} (1997), no.~5, 839--867. \href{https://doi.org/10.2977/prims/1195145020}{doi:10.2977/prims/1195145020}. \bibitem[BK]{BK} A. Berenstein and D. Kazhdan, \emph{Geometric and unipotent crystals II: From unipotent bicrystals to crystal bases}, in \emph{Quantum Groups}, Contemp. Math. \textbf{433}, Amer. Math. Soc., Providence, RI, 2007, 13--88. \href{https://arxiv.org/abs/math/0601391}{arXiv:math/0601391}. \bibitem[BS22]{BS7} R.~Biswal and T.~Scrimshaw, \emph{Kirillov--Reshetikhin crystals $B^{7,s}$ for type $E_7^{(1)}$}, Comm. Algebra \textbf{50} (2022), no.~2, 508--523. \href{https://doi.org/10.1080/00927872.2021.1959923}{doi:10.1080/00927872.2021.1959923}. \bibitem[Cao]{CaoString} P. Cao, \emph{On exchange matrices from string diagrams}, \href{https://arxiv.org/abs/2203.07822}{arXiv:2203.07822} (2022). \bibitem[CM]{CM} V. Chari and A. Moura, \emph{The restricted Kirillov--Reshetikhin modules for the current and twisted current algebras}, Comm. Math. Phys. \textbf{266} (2006), no.~2, 431--454. \href{https://doi.org/10.1007/s00220-006-0032-2}{doi:10.1007/s00220-006-0032-2}. \bibitem[CP]{ChariPressley} V. Chari and A. Pressley, \emph{A Guide to Quantum Groups}, Cambridge University Press, Cambridge, 1994. \bibitem[FeiT]{FeiTheta} J.~Fei, \emph{Crystal structures on theta bases}, preprint, 2026. \bibitem[FeiR]{FeiReddening} J.~Fei, \emph{Reddening sequences for translated Kirillov--Reshetikhin seeds}, accompanying supplement, 2026. \bibitem[FeiS]{FeiSchubert} J.~Fei, \emph{Schubert Cox rings and root operators}, preprint, 2026. \href{https://hexagonmath.org/2611.00668v1}{Hexagon:2611.00668v1}. \bibitem[FKR]{FKR} M. Finkelberg, A. Kuznetsov, L. Rybnikov, and G. Dobrovolska, \emph{Towards a cluster structure on trigonometric zastava}, Selecta Math. (N.S.) \textbf{24} (2018), no.~1, 187--225. \href{https://arxiv.org/abs/1504.05605}{arXiv:1504.05605}. \bibitem[FK]{FK} G. Fourier and D. Kus, \emph{Demazure modules and Weyl modules: the twisted current case}, Trans. Amer. Math. Soc. \textbf{365} (2013), no.~11, 6037--6064. \href{https://doi.org/10.1090/S0002-9947-2013-05846-1}{doi:10.1090/S0002-9947-2013-05846-1}. \bibitem[FL06]{FLTensor} G. Fourier and P. Littelmann, \emph{Tensor product structure of affine Demazure modules and limit constructions}, Nagoya Math. J. \textbf{182} (2006), 171--198. \href{https://arxiv.org/abs/math/0412432}{arXiv:math/0412432}. \bibitem[FL07]{FL} G. Fourier and P. Littelmann, \emph{Weyl modules, Demazure modules, KR-modules, crystals, fusion products and limit constructions}, Adv. Math. \textbf{211} (2007), no.~2, 566--593, \href{https://arxiv.org/abs/math/0509276}{arXiv:math/0509276}. \bibitem[FOS]{FOSPerfect} G. Fourier, M. Okado, and A. Schilling, \emph{Perfectness of Kirillov--Reshetikhin crystals for nonexceptional types}, in \emph{Quantum Affine Algebras, Extended Affine Lie Algebras, and Their Applications}, Contemp. Math. \textbf{506}, Amer. Math. Soc., Providence, RI, 2010, 127--143. \href{https://arxiv.org/abs/0811.1604}{arXiv:0811.1604}. \bibitem[Fra]{Francone} L. Francone, \emph{Cluster structures on Cox rings}, preprint, 2024. \href{https://arxiv.org/abs/2412.04173}{arXiv:2412.04173}. \bibitem[FraL]{FranconeLifting} L. Francone, \emph{The minimal monomial lifting of cluster algebras I: branching problems}, Adv. Math. \textbf{500} (2026), 111102. \href{https://doi.org/10.1016/j.aim.2026.111102}{doi:10.1016/j.aim.2026.111102}. \bibitem[GLS]{GLSFlags} C. Gei\ss, B. Leclerc, and J. Schr\"oer, \emph{Partial flag varieties and preprojective algebras}, Ann. Inst. Fourier (Grenoble) \textbf{58} (2008), no.~3, 825--876. \href{https://doi.org/10.5802/aif.2371}{doi:10.5802/aif.2371}. \bibitem[GY]{GYIntegral} K. R. Goodearl and M. T. Yakimov, \emph{Integral quantum cluster structures}, Duke Math. J. \textbf{170} (2021), no.~6, 1137--1200. \href{https://doi.org/10.1215/00127094-2020-0061}{doi:10.1215/00127094-2020-0061}. \bibitem[GHKK]{GHKK} M.~Gross, P.~Hacking, S.~Keel, and M.~Kontsevich, \emph{Canonical bases for cluster algebras}, J. Amer. Math. Soc. \textbf{31} (2018), no.~2, 497--608. \href{https://doi.org/10.1090/jams/890}{doi:10.1090/jams/890}; \href{https://arxiv.org/abs/1411.1394}{arXiv:1411.1394}. \bibitem[JS26]{JangScrimshaw} I.-S. Jang and T. Scrimshaw, \emph{Special Kirillov--Reshetikhin crystals}, \href{https://arxiv.org/abs/2608.27949}{arXiv:2608.27949} (2026). \bibitem[LNSSS]{LNSSS} C. Lenart, S. Naito, D. Sagaki, A. Schilling, and M. Shimozono, \emph{A uniform model for Kirillov--Reshetikhin crystals II: Alcove model, path model, and $P=X$}, Int. Math. Res. Not. IMRN (2017), no.~14, 4259--4319. \href{https://doi.org/10.1093/imrn/rnw129}{doi:10.1093/imrn/rnw129}. \bibitem[LS]{LenartScrimshaw} C. Lenart and T. Scrimshaw, \emph{On higher level Kirillov--Reshetikhin crystals, Demazure crystals, and related uniform models}, J. Algebra \textbf{539} (2019), 285--304. \href{https://doi.org/10.1016/j.jalgebra.2019.07.036}{doi:10.1016/j.jalgebra.2019.07.036}. \bibitem[JS10]{JS} B.~Jones and A.~Schilling, \emph{Affine structures and a tableau model for $E_6$ crystals}, J. Algebra \textbf{324} (2010), 2512--2542. \href{https://arxiv.org/abs/0909.2442}{arXiv:0909.2442}. \bibitem[Kac]{Kac} V. G. Kac, \emph{Infinite-Dimensional Lie Algebras}, 3rd ed., Cambridge University Press, Cambridge, 1990. \bibitem[KWWY]{KWWY} J. Kamnitzer, B. Webster, A. Weekes, and O. Yacobi, \emph{Yangians and quantizations of slices in the affine Grassmannian}, Algebra Number Theory \textbf{8} (2014), no.~4, 857--893. \href{https://doi.org/10.2140/ant.2014.8.857}{doi:10.2140/ant.2014.8.857}. \bibitem[KK]{KKHead} S.-J. Kang and M. Kashiwara, \emph{Quantized affine algebras and crystals with core}, Comm. Math. Phys. \textbf{195} (1998), no.~3, 725--740. \href{https://doi.org/10.1007/s002200050410}{doi:10.1007/s002200050410}; \href{https://arxiv.org/abs/q-alg/9710008}{arXiv:q-alg/9710008}. \bibitem[KKM92]{KKM} S.-J.~Kang, M.~Kashiwara, K.~C.~Misra, T.~Miwa, T.~Nakashima, and A.~Nakayashiki, \emph{Perfect crystals of quantum affine Lie algebras}, Duke Math. J. \textbf{68} (1992), no.~3, 499--607. \bibitem[KMOY07]{KMOY} M.~Kashiwara, K.~C.~Misra, M.~Okado, and D.~Yamada, \emph{Perfect crystals for $U_q(D_4^{(3)})$}, J. Algebra \textbf{317} (2007), 392--423. \href{https://arxiv.org/abs/math/0610873}{arXiv:math/0610873}. \bibitem[Ked]{KedemPentagon} R. Kedem, \emph{A pentagon of identities, graded tensor products, and the Kirillov--Reshetikhin conjecture}, in \emph{New Trends in Quantum Integrable Systems}, World Sci. Publ., Hackensack, NJ, 2011, 173--193. \href{https://doi.org/10.1142/9789814324373_0010}{doi:10.1142/9789814324373\_0010}; \href{https://arxiv.org/abs/1008.0980}{arXiv:1008.0980}. \bibitem[Kum]{Kumar} S. Kumar, \emph{Kac--Moody Groups, Their Flag Varieties and Representation Theory}, Progr. Math. \textbf{204}, Birkh\"auser, Boston, MA, 2002. \href{https://doi.org/10.1007/978-1-4612-0105-2}{doi:10.1007/978-1-4612-0105-2}. \bibitem[KV]{KV} D. Kus and R. Venkatesh, \emph{Twisted Demazure modules, fusion product decomposition and twisted $Q$-systems}, Represent. Theory \textbf{20} (2016), 94--127. \href{https://doi.org/10.1090/ert/478}{doi:10.1090/ert/478}. \bibitem[Mil]{Milne} J. S. Milne, \emph{Algebraic Groups: The Theory of Group Schemes of Finite Type over a Field}, Cambridge Studies in Advanced Mathematics \textbf{170}, Cambridge University Press, Cambridge, 2017. \href{https://www.jmilne.org/math/Books/iAG2017.pdf}{Author's book text}. \bibitem[Pol]{PolishchukPoisson} A. Polishchuk, \emph{Algebraic geometry of Poisson brackets}, J. Math. Sci. \textbf{84} (1997), no.~5, 1413--1444. \href{https://pages.uoregon.edu/apolish/alg-geom-poisson.pdf}{Author's manuscript}. \bibitem[Pop]{Popov} V. L. Popov, \emph{Picard groups of homogeneous spaces of linear algebraic groups and one-dimensional homogeneous vector bundles}, Math. USSR-Izv. \textbf{8} (1974), no.~2, 301--327. \href{https://doi.org/10.1070/IM1974v008n02ABEH002107}{doi:10.1070/IM1974v008n02ABEH002107}. \bibitem[ST]{ST} A. Schilling and P. Tingley, \emph{Demazure crystals, Kirillov--Reshetikhin crystals, and the energy function}, Electron. J. Combin. \textbf{19} (2012), no.~2, Paper~4. \href{https://arxiv.org/abs/1104.2359}{arXiv:1104.2359}. \bibitem[Sco]{Scott} J. S. Scott, \emph{Grassmannians and cluster algebras}, Proc. London Math. Soc. (3) \textbf{92} (2006), no.~2, 345--380. \href{https://doi.org/10.1112/S0024611505015571}{doi:10.1112/S0024611505015571}. \bibitem[Sta]{Stacks} The Stacks Project Authors, \emph{The Stacks Project}. \href{https://stacks.math.columbia.edu}{stacks.math.columbia.edu}. \end{thebibliography} \end{document}