\documentclass[11pt,reqno]{amsart} \usepackage[a4paper,margin=1.1in]{geometry} \usepackage{amsmath,amssymb,mathtools,booktabs} \usepackage{microtype} \usepackage[hidelinks]{hyperref} \hypersetup{pdftitle={Reddening sequences for translated Kirillov--Reshetikhin seeds}, pdfauthor={Jiarui Fei}} \numberwithin{equation}{section} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \setlength{\emergencystretch}{2em} \title[Reddening sequences for translated KR seeds] {Reddening sequences for translated Kirillov--Reshetikhin seeds} \author{Jiarui Fei} \address{School of Mathematical Sciences, Shanghai Jiao Tong University, Shanghai, China} \email{jiarui@sjtu.edu.cn} \subjclass[2020]{Primary 13F60; Secondary 17B67} \keywords{Reddening sequence, cluster algebra, Kirillov--Reshetikhin module, scattering diagram} \date{} \begin{document} \raggedbottom \setcounter{tocdepth}{2} \begin{abstract} We prove that every untwisted translated Kirillov--Reshetikhin seed admits a reddening sequence. We first find a cluster common to the translated and spectral structures. At a double bond, a strip extension reduces the problem to two boundary degrees. A symmetric covering bounds these degrees by those of an ordinary spectral star. Baxter and Toda recurrences then give the required values in every dimension. Alternating the two extensions proves the local result, and ordinary branches and chamber variables give every fundamental node. The two seeds of type \(G_2\) are treated by explicit maximal green sequences. \end{abstract} \maketitle \section{Reddening for translated seeds} Let \(R=\bigoplus_{s\ge0}R_s\) be the Kirillov--Reshetikhin Cartan algebra at a fundamental node \(r\) of a finite-dimensional simple Lie algebra. Let \(\Sigma_a\), for \(a\ne0\), be its \emph{translated seed}, constructed in \cite[Theorem~5.11]{KRMain}, and let \(\tau\in R_1\) be the canonical Schubert section. We use Bourbaki node numbering, normalize long roots to have squared length two, and write \(t_i=2/(\alpha_i,\alpha_i)\). We use principal coefficients to define the signs of mutable vertices. A vertex is \emph{green} if its coefficient vector is nonnegative and \emph{red} if it is nonpositive. A \emph{reddening sequence} is a finite mutation sequence after which every mutable vertex is red. It is a \emph{maximal green sequence} if each mutation is at a green vertex. For a seed without mutable vertices, the empty sequence is allowed. These are the conventions of \cite[Definition~1.0.4]{COrbBHIT}. \begin{theorem}\label{thm:supp-scope} For every finite simple Lie algebra and every fundamental node, the translated seed \(\Sigma_a\), with \(a\ne0\), admits a reddening sequence. \end{theorem} The principal matrix includes \(\tau\) whenever this vertex is mutable. The resulting affine theta crystals, their degree--level identity, and their simplicity are studied in \cite[Sections~7--9]{KRMain}. Section~\ref{sec:supp-spectral} identifies a common cluster for the translated and spectral structures. In Section~\ref{sec:supp-weighted} we prove reddening for the weighted spectral trees by reducing to a two-strip matrix at the double bond. A covering by an ordinary spectral star supplies the boundary bounds needed for an induction on the strip lengths. Section~\ref{sec:supp-G2} gives the two remaining finite calculations in type \(G_2\) and completes the proof. \section{Comparison with a spectral seed}\label{sec:supp-spectral} \subsection{The simply laced projective bracket} \begin{lemma}\label{lem:next-spectral-tau} At every node of type \(D\) or \(E\), the Schubert section \(\tau\) is, up to scalar, a coordinate of a spectral cluster. Every finite Weyl translate of \(\tau\) has the same property. \end{lemma} \begin{proof} We use the finite opposition convention of \cite[Section~3]{KRMain}, so that \(\tau\) spans the line of weight \(\omega_r\) in \(R_1\). Start with the degree-one spectral frozen section \(c\). From any function already obtained, allow the highest nonzero divided derivative for any current generator \(E_i\) or \(F_i\). Each function so obtained is a homogeneous spectral cluster monomial up to scalar by \cite[Lemma~6.5 and Theorem~1.2]{FeiDESpectral}. A Laurent monomial regular in a polynomial-frozen upper algebra also has nonnegative frozen exponents. These functions are finite-torus weight vectors. Among the weights reached choose \(\nu\) of largest Euclidean norm, and let \(f\) have weight \(\nu\). There are only finitely many weights in \(R_1\). In the rank-one decomposition at \(i\), write \(m=\langle h_i,\nu\rangle\), taking \(h_0=-h_\theta\). The largest highest weight \(l\) of a summand with a nonzero component of \(f\) satisfies \(l\ge |m|\). The two maximal derivatives replace the simple-coroot coordinate \(m\) by \(l\) and \(-l\) and leave its orthogonal projection fixed. Maximality forces \(l=|m|\), so \(f\) is extremal in each simple-root string, and the Tits representative of \(s_i\) carries it, up to scalar, to the opposite maximal derivative. This weight has the same norm. Repeating the argument shows that all Tits translates are reached and remain extremal for every affine simple-root string. Use finite reflections to make \(\nu\) dominant. For every positive finite root \(\beta\) and \(m\in\mathbb Z\), the real root \(\beta+m\delta\) is affine-Weyl-conjugate to a simple root. Its coroot has nonnegative pairing with \(\nu\), since the loop module has level zero. Conjugating this simple-root extremality gives \((e_\beta\otimes t^m)f=0\). Hence the finite weights in the loop submodule generated by \(f\) belong to \(\nu-Q_+\). The module \(R_1\) has essential simple socle \(V(\omega_r)\), and all its weights belong to \(\omega_r-Q_+\), as proved in \cite[proof of Proposition~8.4]{FeiDESpectral}. Thus \(\nu=\omega_r\). Its weight line is unique, so \(f\) is a nonzero multiple of \(\tau\). A nonconstant cluster monomial of degree one is a single coordinate. The same extremal-weight argument yields its finite Weyl translates. \end{proof} \begin{lemma}\label{lem:next-projective-poisson} At \(a=1\) the spectral cluster structure admits a compatible Poisson bracket whose projectivization is the translated trigonometric Poisson bracket. We allow simultaneous reversal of the signs of the bracket and the exchange matrix. \end{lemma} \begin{proof} Let \(Y=\operatorname{Proj}R\). All statements about projectivization refer to the bracket on the degree-zero function field. Write \(P\) for the spectral bracket after the grading twist of \cite[Proposition~6.11 and Appendix~B, ``The compatible grading twist'']{FeiDESpectral}. Its proof starts with \(B\Lambda=-(I\mid0)\) and adds a bivector in the exterior square of the grading torus. Since the grading directions lie in \(\ker B\), this operation leaves the compatibility matrix unchanged. The result satisfies \[ \mathcal L_{D_i}P=(\sigma/2)D_i\wedge H_i \qquad(D_i=E_i,F_i;\ i>0),\qquad \sigma\in\{1,-1\}. \] Multiply \(P\) and its exchange matrix by \(-1\) together if necessary so that \(\sigma=-1\). The translated projective bracket \(\overline \Pi_1\) satisfies the identical finite-root equations, and its affine equation has the same coefficient, by \cite[Lemma~6.4, proof]{KRMain}. Write bars for projections to \(Y\). The bivector \(Q=\overline P-\overline\Pi_1\) is invariant under the finite group \(G\). The spectral structure has an adapted chart for \(D=F_0=e_\theta\otimes(1+u)^{-1}\) at \(b_q\), where \(q=\pi^{-1}(0)\), by \cite[Theorem~1.2]{FeiDESpectral}. We compute the Lie derivative in this chart. In seed coordinates \(z\), let \(D=cz^v\partial_{b_q}\), with \(v\) supported on mutable variables and \(v_{\rm mut}\) equal to the \(b_q\)-column of \(B\) up to its common sign. For \(j\ne b_q\) its logarithmic coefficient is \(a_j=\Lambda_{b_qj}-\sum_l v_l\Lambda_{lj}\). Direct differentiation gives \(\mathcal L_D P=D\wedge A\) with \(A(z_j)=a_jz_j\); the coefficient \(a_{b_q}\) can be chosen freely. In fact, writing \(a=(e_{b_q}-v)\Lambda\) first, one has \(Ba^{\mathsf T}=-v_{\rm mut}^{\mathsf T}\) by compatibility. Changing \(a_{b_q}\) by the appropriate sign gives \(Ba^{\mathsf T}=0\). The section degree and the finite Cartan span this kernel by \cite[Section~6, ``Finite root fields and the compatible bracket'']{FeiDESpectral}. Consequently \(A\) is a constant linear combination of these fields. Upon passage to \(Y\) the Euler field disappears. Subtracting the translated affine covariance thus gives \[ \mathcal L_{\overline F_0}Q=\overline F_0\wedge\overline H \qquad\text{for some constant }H\in\mathfrak h. \] If \(i>0\) and \(i\ne q\), then \([E_i,F_0]=0\). Differentiate this last identity by \(E_i\) and use the finite invariance of \(Q\). It follows that \(\alpha_i(H)\,\overline F_0\wedge\overline E_i=0\). These two projective vector fields are independent. Indeed, on the cone these two derivations have distinct active frozen sections \(b_q\) and \(b_i\), and each annihilates the other active section. They also annihilate a third frozen section of positive degree. In a rational relation among \(F_0,E_i\) and the Euler field, that third section forces the Euler coefficient to vanish. The two active sections then force the other coefficients to vanish. Hence their projective vector fields are independent. If \(q=0\) this settles every finite simple root. Otherwise \(q\) is minuscule and, in types \(D,E\), it differs from the finite node joined to \(0\) in the affine diagram. Thus \([F_q,F_0]=0\). The frozen sections active for these lowering operators, \(b_{\pi^{-1}(q)}\) and \(b_q\), are distinct; the same argument with three frozen sections applies. Differentiation by \(F_q\) now gives \(\alpha_q(H)=0\) as well. All finite simple roots vanish on \(H\), so \(H=0\). It follows that \(Q\) is invariant under \(\mathfrak g\) and under \(e_\theta\otimes t^{-1}\), where \(t=1+u\). Taking finite brackets gives \(\mathfrak g\otimes t^{-1}\), since the adjoint module is simple. Further brackets give \(\mathfrak g\otimes t^{-j}\) for every \(j\ge1\). In the finite jet algebra \(\mathbb C[u]/(u^N)\) through which the action factors, \(t^{-1}-1=-u/(1+u)\) generates the maximal ideal and has invertible linear term. Its powers together with \(1\) span the jet algebra. The acting jet group is connected, with connected reductive quotient and unipotent kernel. The infinitesimal invariance therefore makes \(Q\) invariant on the open spherical orbit \(\operatorname{Gr}^\lambda\subset Y\). Its regular locus is a nonempty invariant open subset of this transitive orbit, hence the whole orbit. At the coweight point the finite-torus weights of the tangent space are \(\beta\), each with multiplicity \(\langle\beta,\lambda\rangle\), for positive finite roots \(\beta\) (simultaneous negation gives the opposite convention). This is the root-space description of the tangent quotient \(\mathfrak g[[u]]/(\mathfrak g[[u]]\cap \operatorname{Ad}_{u^\lambda}\mathfrak g[[u]])\). No pair of these weights sums to zero. Hence the exterior square of the tangent space has no torus-fixed vector. Thus \(Q\) vanishes on the orbit and consequently in the function field. \end{proof} \subsection{The projective bracket at a weighted node} \begin{lemma}\label{lem:weighted-common-tau} In the spectral structure of \cite{FeiBCFG}, the Schubert section \(\tau\) is a cluster coordinate up to scalar at every fundamental node. \end{lemma} \begin{proof} Start with the frozen component section \(c\) of degree \(L=t_r\), and apply arbitrary successive highest nonzero divided derivatives for the shifted current generators. By the adapted charts and the maximal-derivative theorem of \cite{FeiBCFG}, each result is a spectral cluster monomial of degree \(L\). Apply the argument of Lemma~\ref{lem:next-spectral-tau} in \(R_L\). Its essential simple socle has highest weight \(L\omega_r\), and all its weights lie in \(L\omega_r-Q_+\), by \cite[proof of Theorem~9.2]{KRMain}. The maximal-norm argument therefore reaches weight \(L\omega_r\). That extremal line is one-dimensional, spanned by \(\tau^L\). Thus \(\tau^L\) is a scalar multiple of a spectral cluster monomial. The algebra is factorial and its seed coordinates and \(\tau\) are prime. Unique factorization places \(\tau\) in that seed, also when \(L>1\). \end{proof} \begin{lemma}\label{lem:weighted-projective-poisson} In each non-evaluation fundamental component of types \(B,C,F,G\) the spectral seed of \cite{FeiBCFG} has a compatible Poisson bracket whose projective restriction is the translated trigonometric bracket at \(a=1\). \end{lemma} \begin{proof} Write \(R\) for the Cartan algebra and \(Y=\operatorname{Proj}R\). Let \(\mathfrak t=\mathfrak h\oplus\mathbb C\mathsf E\) denote the finite Cartan together with the degree derivation. All fields in \(\mathfrak t\) are diagonal in every spectral seed. Their eigenvalue vectors span \(\ker B\) by \cite[section ``Poisson brackets and boundary derivations'']{FeiBCFG}. In particular the corresponding rational vector fields are linearly independent over \(\operatorname{Frac}R\): on the dense seed torus this is the full column rank of their constant eigenvalue matrix. Use the sign of the spectral exchange matrix for which the raising charts have nonnegative active columns. Choose a compatible skew form with \(B\Lambda=-(\operatorname{diag}(\delta_v^{-1})\mid0)\); full row rank and skew-symmetrizability give such a form. Write \(P\) for its log-canonical Poisson bracket. Consider an adapted spectral chart for a generator \(D\) of color \(i\) at its active frozen \(b\). If \(D=cz^v\partial_b\), with \(v_b=0\), the spectral current theorem gives \[ v_j=\epsilon\,\delta_j B_{jb}/t_i\quad(j\text{ mutable}), \qquad \epsilon=1\text{ for }E_i,\quad \epsilon=-1\text{ for }F_i. \] Here \(z^v\) may include other frozen variables. From the values on coordinate pairs one obtains \(\mathcal L_DP=D\wedge A_D\). The diagonal coefficient vector of \(A_D\) may be taken to be \(a=(e_b-v)\Lambda+(\epsilon/t_i)e_b\). Indeed the final summand does not affect \(D\wedge A_D\), and \(B((e_b-v)\Lambda)^{\mathsf T}=-\operatorname{diag}(\delta_j^{-1}) v_{\rm mut}^{\mathsf T}\). Hence \(Ba^{\mathsf T}=0\) and \(A_D\) is a constant element of \(\mathfrak t\). The eigenvalue in \([A_D,D]\) is \((v-e_b)a^{\mathsf T}=-\epsilon/t_i\) by skewness. For finite \(i\) write \(A_i^+=A_{E_i}\), \(A_i^-=A_{F_i}\). With \(\alpha_i\) interpreted as the finite character extended by zero on \(\mathsf E\), this says \(\alpha_i(A_i^+)=\alpha_i(A_i^-)=-1/t_i\). We normalize the finite covariances simultaneously, including across a multiple bond. The commutator \([E_i,F_i]=H_i\) preserves \(P\) since \(H_i\) is diagonal. Its Lie derivative, computed from the two simple covariances, gives \[ H_i\wedge(A_i^++A_i^-) +\bigl(\alpha_i(A_i^-)-\alpha_i(A_i^+)\bigr)E_i\wedge F_i=0. \] Independence of the torus fields and \(\alpha_i(H_i)=2\) show that \(A_i^++A_i^-=-H_i/t_i\). If \(i\ne j\) are finite, the spectral frozen sections active for \(E_i\) and \(F_j\) are different. This follows from the component table of \cite{FeiBCFG}: the components under consideration are neutral except for odd \(B\), where \(\pi=(0\ 1)\). Thus \(E_i\wedge F_j\ne0\), as can be seen by evaluation on their two active sections. Differentiating \([E_i,F_j]=0\) yields \(\alpha_i(A_j^-)=\alpha_j(A_i^+)\). Consequently \[ \alpha_i(A_j^+)+\alpha_j(A_i^+)=-\alpha_i(H_j)/t_j =-(\alpha_i,\alpha_j). \] The same equality holds for \(i=j\). Define \(C_i=-H_i/(2t_i)-A_i^+\). These assignments satisfy \(\alpha_i(C_j)+\alpha_j(C_i)=0\) for all \(i,j\). Since the finite simple-root characters are independent in \(\mathfrak t^*\), the maps \(\alpha_i\mapsto C_i\) extend to a skew map \(\mathfrak t^*\to\mathfrak t\). Adding the corresponding grading bivector, with the contraction sign chosen so that the raising correction is \(C_i\), gives a Poisson bracket \(P'\) with \[ \mathcal L_D P'=D\wedge\bigl(-H_i/(2t_i)\bigr), \qquad D=E_i,F_i,\quad i>0. \] The lowering correction is \(-C_i\) because \(F_i\) has the opposite weight. The twist preserves compatibility and the Poisson identity, since the commuting torus fields preserve \(P\). On \(Y\) the translated bracket \(\overline\Pi_1\) obeys these same finite covariance equations, and obeys the affine simple covariance, by \cite[Lemma~6.4, proof]{KRMain}. Let \(Q=\overline P'-\overline\Pi_1\). It is finite-\(G\)-invariant, and the adapted spectral chart for \(F_0\) gives \(\mathcal L_{\overline F_0}Q=\overline F_0\wedge\overline H\) for some constant \(H\in\mathfrak h\) by the preceding computation for diagonal fields. Euler fields disappear on projectivization. If \(i>0\) and \(i\ne q=\pi^{-1}(0)\), then \([E_i,F_0]=0\) and their spectral active frozen sections differ. Their projective wedge is nonzero: on the cone a third frozen section annihilated by both has positive degree, excluding an Euler component in any rational linear dependence. Differentiating the last identity by \(E_i\) therefore gives \(\alpha_i(H)=0\). If \(q=0\), these equations cover all finite simple roots. In odd \(B\) we have \(q=1\). In the non-evaluation cases at issue the finite node \(1\) is not adjacent to \(0\), so \([F_1,F_0]=0\), and their active spectral sections are different. Using \(F_1\) gives \(\alpha_1(H)=0\) as well. Thus \(H=0\). We now apply the final argument of Lemma~\ref{lem:next-projective-poisson}. The finite current algebra and \(e_\theta\otimes(1+u)^{-1}\) generate the acting jet algebra, so \(Q\) is invariant on the open spherical orbit. At the coweight point its tangent weights are positive finite roots, with multiplicities \(\langle\beta,\omega_r^\vee\rangle\), up to a common sign. Their pairwise sums are nonzero. Hence the invariant regular bivector \(Q\) vanishes, proving the equality of brackets. \end{proof} \subsection{A common cluster for the spectral and translated structures} Write \(L=2/(\alpha_r,\alpha_r)\), let \(k_i\) be the affine comarks, and let \(M=\langle\theta,\omega_r^\vee\rangle\). We use the spectral structure of \cite{FeiDESpectral} in types \(D,E\) and that of \cite{FeiBCFG} in types \(B,C,F,G\). At parameter \(a\ne0\), the spectral structure means the transport of its parameter-one structure by loop rotation, as specified at the end of the proof. \begin{theorem}\label{thm:next-common-spectral-seed} Suppose \(M>1\). At each nonzero parameter the spectral and translated structures have a common cluster up to rescaling, with the same frozen sections. One principal matrix admits a reddening sequence if and only if the other does. \end{theorem} We first prove two lemmas about frozen variables. We use exchange matrices indexed by mutable rows. \begin{lemma}\label{lem:next-normal-calibration} Every Poisson-normal function on the standard Schubert cell \(U(w)\) is a polynomial in the terminal minors \(f_i=D_{\Lambda_i,w\Lambda_i}\). \end{lemma} \begin{proof} Recall that \(p\) is \emph{Poisson-normal} if \(\{p,S\}\subset pS\), where \(S=\mathbb C[U(w)]\). Let \(p\ne0\) have this property. In the Laurent ring of a word-cluster chart \(x\), write \(\{p,x_j\}/x_j=pq_j\). The Laurent polynomial on the left has support contained in that of \(p\). If it is nonzero, the Newton polytope of \(pq_j\) is the sum of the Newton polytopes of \(p\) and \(q_j\); hence \(q_j\) is constant. Denote the constants by \(c_j\), allowing zero. Every exponent \(v\) of \(p\) satisfies \(v\mathsf P=c\), where \(\mathsf P\) is the log-bracket matrix. The same argument applies to \(x_k'\) in the adjacent chart. Since \(x_kx_k'=M_++M_-\) also has constant logarithmic bracket with \(p\), the distinct monomials \(M_+,M_-\) have the same logarithmic bracket with \(p\). Thus \(B_kc^{\mathsf T}=0\) for every mutable row. Compatibility gives \(v_{\rm mut}=0\). The mixed Laurent ring is polynomial in the frozen variables, proving the assertion. \end{proof} \begin{lemma}\label{lem:next-project-optimization} If a frozen column has a reachable nonnegative chart, it still has one after restricting to any subset \(I\) of the mutable positions of a prescribed starting seed. The same holds for nonpositive charts. \end{lemma} \begin{proof} Write \(B\) for the starting principal matrix, \(z\) for the column and \(z_t\ge0\) for its value at a target chart. Transport from \(t\) back to the start \(t_0\). In the usual convention with frozen rows the exchange matrix is \(B_t^{\mathsf T}\). Sign coherence makes transport linear on nonnegative combinations of the principal coefficient rows. Thus tropical duality gives \[ z=\bigl(C_{t_0}^{\,B_t^{\mathsf T};t}\bigr)^{\mathsf T}z_t =G_t^{\,B;t_0}z_t \] by \cite[(1.13)]{NZTropical}. Conversely \(G_t^{B;t_0}v\) reaches \(v\) for every \(v\ge0\), by invertibility of coefficient mutation. The enough \(g\)-pairs theorem gives an \(I\)-mutation path to \(t'\) for which \((G_t)_{I,*}=(G_{t'})_{I,I}Q\), \(Q\ge0\) \cite[Proposition~4.6 and Theorem~4.8]{CaoLiGPairs}. The submatrix \((G_{t'})_{I,I}\) is the \(g\)-matrix of the restricted pattern: in the backward principal coefficient formula above, the \(I\)-by-\(I\) submatrix evolves entirely on \(B_{I,I}\) along an \(I\)-path. Hence \(z_I\) reaches \(Qz_t\ge0\) in the restricted pattern. Apply the argument to \((-B,-z)\) for the other sign. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:next-common-spectral-seed}] Work first at \(a=1\). Lemmas~\ref{lem:next-projective-poisson} and \ref{lem:weighted-projective-poisson} identify the two projective brackets. Lemmas~\ref{lem:next-spectral-tau} and \ref{lem:weighted-common-tau} place \(\tau\) in a spectral cluster. It is mutable: each spectral frozen section of finite color has a nonzero finite raising derivative, whereas every finite raising operator kills \(\tau\). The remaining frozen section has nonzero \(F_{\pi(0)}\)-derivative, and \(F_{\pi(0)}\tau=0\) as explained below. Divide a spectral frozen section \(b_i^{\rm sp}\) by \(\tau^{Lk_i}\) and remove the source translation. The result \[ p_i=\Psi_1^{-1}(b_i^{\rm sp}/\tau^{Lk_i})\in S=(R[\tau^{-1}])_0=\mathbb C[U(w)] \] is Poisson-normal for the standard bracket. The logarithmic Hamiltonian of a spectral frozen section preserves the upper algebra; after inverting \(\tau\), so does the logarithmic Hamiltonian of \(\tau\). Subtracting \(Lk_i\) times the latter and taking degree zero gives the assertion by projective Poisson comparison. Lemma~\ref{lem:next-normal-calibration} gives \(p_i\in\mathbb C[f_0,\ldots,f_n]\). Every finite \(E_j\) fixes \(\tau\), as does \(F_{\pi(0)}\): if \(\pi(0)\) is finite it has mark one and differs from \(r\); the affine \(F_0\) fixes the finite highest section. These derivations thus descend to \(S\) after conjugation by \(\Psi_1^{-1}\). Their active colors on the \(p_i\) and the \(f_i\) agree by the respective current theorems, including the unconditional lowering charts of the translated structure. For \(i>0\), apply \(E_j\) for all finite \(j\ne i\). Algebraic independence of the terminal minors gives \(p_i\in\mathbb C[f_i,f_0]\). Applying \(F_{\pi(0)}\), which acts on \(f_0\) alone among the terminals, gives \(p_i\in\mathbb C[f_i]\). Both sections have \(E_i\)-string length one, so this polynomial is linear. For \(i=0\) the finite raising operators give \(p_0\in\mathbb C[f_0]\); the \(F_{\pi(0)}\)-strings give linearity. Both \(p_i\) and \(f_i\) are finite-torus eigenvectors. Each \(f_i\) has nonzero weight, since it is nonconstant and every cell root has strictly positive finite part. Thus the constant term vanishes. Restoring translation and the power of \(\tau\) identifies the frozen sections up to scalar. In a spectral seed containing \(\tau\), freeze this coordinate and replace every other \(x\) by \(\Psi_1^{-1}(x/\tau^{\deg x})\). The resulting seed \(\Gamma\) has polynomial frozen variables \(f_i\) and mixed upper algebra \(S\). Its mixed charts are the degree-zero localizations of the original charts with \(\tau\) inverted. Every coordinate in them is a prime of \(S\), since the original coordinate is prime and distinct from \(\tau\), and \(R[\tau^{-1}]=S[\tau^{\pm1}]\). At any height-one prime of \(S\) the initial mixed chart is defined unless its mutable coordinate vanishes, in which case its adjacent chart is defined there. Normality now gives the upper-algebra equality. The matrix \(\bar B\) deletes the row and column of \(\tau\); it has full row rank since homogeneity expresses the deleted column in terms of the others. For the matrix \(A\) with rows \(e_x-\deg(x)e_\tau\), the remaining rows of the old matrix equal \(\bar BA\), and the new log-bracket matrix is \(A\mathsf PA^{\mathsf T}\). Thus compatibility preserves the mutable factors \(\delta_j^{-1}\). Each original frozen coordinate has reachable charts of both signs by the spectral current theorem. Lemma~\ref{lem:next-project-optimization} therefore gives a nonnegative chart in \(\Gamma\) for each \(f_i\), in the normalization \(\bar B\overline{\mathsf P} =-(\operatorname{diag}(\delta_j^{-1})\mid0)\). The finite Cartan acts diagonally throughout these charts. So does loop rotation: the Schubert logarithmic Hamiltonian of \(f_i\) is a finite Cartan field minus \(k_i\) times rotation, and \(f_i\) is log-canonical there. The corresponding \(n+1\) gradings span \(\ker\bar B\). Indeed the finite roots occurring on the cell span the finite root space: their span contains \(\alpha_r\) and is stable under each \(s_j\), \(j\ne r\), hence contains all simple roots by connectedness. The two cell roots \(\theta,\theta+\delta\) give the remaining direction since \(M>1\). Let \(e_i\) be the ordinary left simple-root derivation on \(S\). For \(i\ne r\), assign degree zero to \(\alpha_i\) and degree one to every other affine simple root, with the coordinate weights taken in the positive inversion cone. Every cell root has positive degree: \(\alpha_r\) is the only simple inversion of \(w\). Thus \(S\) is connected and positively graded and \(e_i\) has degree zero. The fundamental string gives \(e_i f_i\ne0,\ e_i^2 f_i=0\). The Schubert projective covariance \(\mathcal L_{e_i}\Pi=e_i\wedge(-H_i/(2t_i))\) gives the root Hamiltonian identity with coefficient \(t_i>0\) by \cite[Corollary~4.6]{KRMain}. Hamiltonian recognition \cite[Proposition~4.5]{KRMain} makes the nonnegative chart at \(f_i\) adapted to \(e_i\). This applies in particular to the ordinary positive affine root \(e_0\). For \(i=r\), use the spectral \(E_r\)-chart directly. Write its coefficient as \(N\), and let \(P\) be its mutable support. Since \(E_r\tau=0\), the highest divided derivative of the cluster monomial \(\tau b_r^{\rm sp}\) is \(\tau N\). It is a spectral cluster monomial, so \(\tau\) occurs together with all members of \(P\). By \cite[Theorem~6.2]{CaoLiGPairs} this chart is connected to the adapted chart by mutations avoiding \(P\). Those mutations preserve the \(b_r^{\rm sp}\)-column and their exchanges are independent of \(b_r^{\rm sp}\). The chart containing \(\tau\) is still adapted. All charts containing \(\tau\) are connected with \(\tau\) fixed by the same theorem. Since \(E_r\) commutes with source translation, its dehomogenization gives an \(e_r\)-chart in \(\Gamma\). Choose a reduced word \(w=s_{i_1}\cdots s_{i_d}\) and write \(w_{y\) and the same \(x\), has already occurred. Consequently \(L_y,A_x\) are adjacent at this moment, with precisely \(x+y\) letters before them. Their difference is \(c_{xy}\). It remains to check that the signed basis mutation agrees with this adjacent interchange. Let \(F(L_v)=-\Omega(v,y)\) and \(F(A_u)=-\Omega(u,x)\). The pairing of any consecutive difference \(D-D'\) with \(c_{xy}\) is \(F(D)-F(D')\). In the current ordered list the nonzero values of \(F\) form a consecutive string of \(+1\) values immediately before \(L_y,A_x\) and a consecutive string of \(-1\) values immediately after them. Here is a verification that no zero can split either string. Set \(k_x=1\) for even \(x\) and \(k_x=2\) for odd \(x\), and similarly \(k_y\). The positive letters are the existing letters with indices \(x-k_x,\ldots,x-1\) or \(y-k_y,\ldots,y-1\). The only potentially borderline comparisons between a zero and a positive letter are governed by the cells \[ (x-k_x-1,y-k_y),\qquad(x-k_x,y-k_y-1). \] Their diagonals differ from \(d=x-y\) by \(-k_x-1+k_y\) and \(-k_x+k_y+1\) respectively. The first is strictly earlier and the second strictly later, except in the two cases below: \[ \begin{array}{c|c|c} \text{comparison}&\text{equal-diagonal case}&\text{tie order}\\\hline \text{zero }A\text{ before positive }L&x\text{ even},y\text{ odd} &x-2\text{ even precedes }x\\ \text{positive }A\text{ after zero }L&x\text{ odd},y\text{ even} &x-2\text{ odd follows }x \end{array} \] The same check for negative letters uses cells \((x+k_x,y+k_y+1)\) and \((x+k_x+1,y+k_y)\): the equal-diagonal exceptions are, respectively, odd \(x\) with \(x+2\) earlier, and even \(x\) with \(x+2\) later. All other comparisons are strict diagonal comparisons. Missing boundary letters are simply omitted. This proves the assertion about consecutive values. Thus the positive part of the exchange-matrix column at the mutated vertex has entry one at each existing adjacent position and zero elsewhere. Its mutation negates \(L_y-A_x\) and adds it to the adjacent differences, exactly as the interchange does. Induction proves \eqref{eq:rectangle-c-root} and its greenness. The final list is \(A_0,\ldots,A_X,L_0,\ldots,L_Y\), which also proves the usual Baxter terminal identity: the strip directions are preserved up to permutation and the arrow direction is replaced by \(-e-u_X-v_0\). For the star, the additional arrow and strip directions do not change this calculation. During the first Baxter word the second arrow is \(p_{2,b}-p_{0,1}\); its pairing with \(c_{xy}\) is \(-\delta_{x,1}\), so its basis vector never changes. The unused leaf strip is orthogonal to every root. During the second Baxter word the reversed first arrow is \(p_{0,a}-p_{1,1}\); its pairing with the root of the second rectangle is \(-\delta_{x,X-1}\), again nonpositive. Thus the second word starts with the same central-strip and second-arrow basis used in \eqref{eq:rectangle-c-root}. \subsubsection{The boundary recurrence} During green mutations the lattice vector of the auxiliary vertex changes by \(z\mapsto z+[\langle z,c\rangle]_+c\). Write \(b_{xy}\) for its pairing with the vertex mutated at cell \((x,y)\), and \(q_{xy}=[b_{xy}]_+\). If its incoming pairing with \(c_{xy}\) is \(f_{xy}\), then \eqref{eq:rectangle-root-pairing} gives \begin{equation}\label{eq:scalar-rectangle-recurrence} b_{xy}=f_{xy}+\sum_{(u,v)\prec(x,y)}q_{uv} \bigl(\Omega(u,x)-\Omega(v,y)\bigr), \qquad q_{xy}=[b_{xy}]_+. \end{equation} Here \(\prec\) is exactly the diagonal and parity order above. The resulting contribution to \(H\) is \(\sum q_{xy}c_{xy}\). This triangular scalar recurrence uniquely determines the pairing at every mutation. \subsubsection{The long-strip coefficient for \((m,m,m)\)} Set \(X=Y=2m-2\). For the first arrow, the incoming row \(e_{S_{m-1}}\) gives \(f_{xy}=\delta_{x,X}\). The exact solution of \eqref{eq:scalar-rectangle-recurrence} is \begin{equation}\label{eq:long-first-pivots} b_{xy}=q_{xy}= \begin{cases} 1,&(x,y)=(X,X),\\ 1,&(x,y)=(2r+1,2r),\quad0\le r\le m-2,\\ 0,&\text{otherwise}. \end{cases} \end{equation} We verify the formula in the recurrence. Before diagonal \(0\), no indicated cell has occurred and \(x=X\) is impossible. On diagonal \(0\), the first cell with nonzero pairing is \((X,X)\); subsequent cells have \(x=y\) and their two \(\Omega\) terms cancel. On diagonal \(1\), the even-\(x\) part gives zero: only \((X,X-1)\) can see the boundary term, and its value is \(1-1=0\). The odd-\(x\) part is read from \(x=X-1\) down to \(x=1\). Each cell obtains value one from the immediately preceding indicated cell; all column terms vanish since their indices are even. After diagonal \(1\), all indicated cells have occurred. The sums then reduce to \[ \delta_{x,X}+ \sum_{u\in\{1,3,\ldots,X-1,X\}}\Omega(u,x) -\sum_{v\in\{0,2,\ldots,X\}}\Omega(v,y) =\delta_{x,0}+\delta_{x,1}. \] A remaining cell has \(x-y\ge2\) and hence \(x\ge2\), so this is zero. This proves every case of \eqref{eq:long-first-pivots}. The coefficient of \(T_{m-1}\) in \(c_{xy}\) is one exactly when \(x\in\{X-1,X\}\). Thus the first word contributes exactly two to the required coefficient. For the second arrow the roots of the first rectangle share only their \(\lambda\) components with the roots of the second rectangle. Consequently its initial pairing is \[ f_{xy}=\delta_{x,X}+ \sum_{u\in\{1,3,\ldots,X-1,X\}}\Omega(u,x) =\delta_{x,0}+\delta_{x,1}. \] The exact pairing values for the second rectangle are \begin{equation}\label{eq:long-second-pivots} b_{xy}= \begin{cases} 1,&x=0,\ y\text{ even},\\ 0,&x=0,\ y\text{ odd},\\ -(m-1-\lfloor y/2\rfloor),&x=1,\\ 0,&x\ge2. \end{cases} \end{equation} Indeed the proposed cells with positive pairing are just \((0,2r)\), \(0\le r\le m-1\), read in decreasing \(r\). At \(x=0\) their adjacent column terms alternate between zero and minus one, giving the first two cases. At \(x=1\) and \(y\in\{2s,2s+1\}\), exactly \(m-s\) of these cells have already been read; the row contribution is \(-(m-s)\) and the column contribution is zero, so \(b=1-(m-s)\). At \(x\ge2\) the row contribution is zero. If \(y\) is odd, both neighboring even columns have already occurred and their column contributions cancel; if \(y\) is even all column contributions vanish. This proves \eqref{eq:long-second-pivots}. Every cell with positive pairing has \(x=0\), so none contributes to the long-strip coefficient, even when \(m=2\). We also need the initial exchange vector for the Toda part. Summing the vectors \(c_{xy}\) from \eqref{eq:long-first-pivots} and the positive part of \eqref{eq:long-second-pivots}, the added vector is \[ K=-\sum_{r=1}^{m}p_{0,r}-(m-1)p_{0,1}-p_{0,m} +x_{0,1}-x_{0,m} +\sum_{r=1}^{m}p_{1,r}+\sum_{r=1}^{m}p_{2,r}. \] Pairing \(z+K\) with the strip directions gives precisely \begin{equation}\label{eq:long-Toda-input} f_{\mathrm{Toda}}=(m-1)e_{s_{0,1}}-m e_{t_{0,1}}. \end{equation} There are no entries on the other strips or the reversed arrows. To check the latter, a reversed arrow is \(p_{0,m}-p_{\ell,1}\): the pairing of \(K\) with it is \(1\), while the original \(z\) pairs with it by \(-1\). \subsubsection{The short-strip coefficients for \((m+1,m,m)\)} Here \(X=2m\) and \(Y=2m-2\). For the first arrow, \(f_{xy}=\delta_{y,0}\). The exact solution is \begin{equation}\label{eq:short-first-pivots} b_{xy}=q_{xy}= \begin{cases} 1,&(x,y)=(0,0),\\ 1,&(x,y)=(2r,2r-1),\quad1\le r\le m-1,\\ 1,&(x,y)=(X,Y),\\ 0,&\text{otherwise}. \end{cases} \end{equation} Before diagonal \(0\) there are no contributions. On diagonal \(0\) only \((0,0)\) contributes, because all later cells have equal row and column indices. On diagonal \(1\) the even-\(x\) cells are read upwards and each gets value one from the preceding indicated cell. The odd-\(x\) cells then give zero. After this diagonal, the sum over the indicated points already read reduces to \[ b_{xy}=-\delta_{x,Y+1}+\delta_{y,Y-1}+\delta_{y,Y}. \] On diagonal \(2\) this vanishes except at \((X,Y)\), where it is one; the possible pair \((Y+1,Y-1)\) has its two terms cancel. Adding the final cell \((X,Y)\) changes this formula to \(\delta_{y,Y}\), since \(\Omega(X,x)=\delta_{x,Y+1}\) and \(\Omega(Y,y)=\delta_{y,Y-1}\) on the relevant ranges. Every later cell has \(x-y\ge3\) and hence \(y\le Y-1\), so its value is zero. This verifies all cases, including \(m=2\). The coefficient of \(t_1\) in \(c_{xy}\) is one exactly when \(y\in\{0,1\}\). Exactly \((0,0)\) and \((2,1)\) from \eqref{eq:short-first-pivots} contribute, so \(H_{t_1}=2\). All coefficients on the other leaf remain zero. The sum of the indicated vectors \(c_{xy}\) is \[ K=-\sum_{r=1}^{m+1}p_{0,r}+\sum_{r=1}^{m}p_{1,r} +p_{1,1}+x_{1,1}-x_{1,m}. \] Pairing \(z+K\) with the relabelled basis after the first Baxter word gives exactly \begin{equation}\label{eq:short-after-Baxter} f'=-e_{t_{1,m-1}}. \end{equation} For example the extra \(p_{1,1}\) cancels the original pairing with \(s_{1,1}\), and the two \(x\) terms leave only the pairing with \(t_{1,m-1}\). The total momenta pair trivially with strip directions, and direct pairing with the reversed first arrow and the unchanged second arrow gives zero. The second Baxter word only has roots on colors \(0,2\) and their arrow, so all its pairings with the auxiliary vertex vanish. The \(c\)-vector at every subsequent Toda mutation is a nonnegative combination of strip directions. Its pairing with \eqref{eq:short-after-Baxter} is nonpositive, so these mutations leave the auxiliary coefficient vector unchanged. This proves both short-strip identities. \subsubsection{The Toda calculation for the long-strip coefficient} On a strip of dimension \(m\), write \(s_h,t_h\) for \(1\le h\le m-1\). Its principal matrix satisfies \[ B_{s,s}=B_{t,t}=0,\qquad B_{s_h,t_k}=2\delta_{hk} -\delta_{h,k-1}-\delta_{h,k+1}. \] Each round of mutations at all \(s\)-vertices or all \(t\)-vertices negates this matrix. Let \(r\) count rounds from zero, and let \(c_{r,h}\) be the coefficient vector at height \(h\) in round \(r\). With zero boundary values at \(h=0,m\), these vectors satisfy \[ c_{0,h}=s_h,\qquad c_{1,h}=t_h+s_{h-1}+s_{h+1},\qquad c_{r+1,h}=c_{r,h-1}+c_{r,h+1}-c_{r-1,h}. \] To prove that these mutations are green, let \(A\) be the adjacency matrix of the path \(1,\ldots,m-1\) and define \(V_{-1}=0\), \(V_0=I\), \(V_{r+1}=AV_r-V_{r-1}\). For \(0\le r\le m-1\) the entries are \[ (V_r)_{hk}=1 \quad\Longleftrightarrow\quad |h-k|\le r\le\min(h+k-2,\,2m-h-k-2),\quad r\equiv h-k\pmod2, \] and are zero otherwise. These indicators satisfy the same recurrence, including the zero boundary values, as is seen by splitting according to the two ends of the interval. In particular they are nonnegative, \(V_{m-1}=0\), and \(V_{m-2}\) is the reversal permutation. Hence \(c_r=V_rs+V_{r-1}t\) is nonnegative and gives the actual mutation vectors by induction. At height one this simplifies to \[ c_{r,1}=s_{r+1}+t_r,\qquad0\le r\le m-1, \quad t_0=s_m=0. \] Starting with the row \eqref{eq:long-Toda-input}, an elementary induction shows that before round \(r\), for \(0\le r\le m-1\), its entry on the kind of vertex mutated in that round at height one is \(m-1-r\) and its entry on the other kind of vertex at height one is \(-(m-r)\); all entries at higher heights are zero. For \(r\le m-2\) only the height-one active vertex has positive pairing with the auxiliary vertex. Mutation there negates \(m-1-r\) and changes the other entry to \(-(m-r)+2(m-1-r)=m-2-r\). The off-diagonal strip entries are \(-1\), so they produce no positive contribution. Thus this is the auxiliary exchange vector for the next Toda round. At the final round the active entry is zero and no positive pairing remains. The positive roots that can add to \(H\) are therefore exactly \(c_{r,1}=s_{r+1}+t_r\) for \(0\le r\le m-2\), with weights \(m-1-r\). None contains \(t_{m-1}\). All other strips are skew-orthogonal and their auxiliary coefficient vectors stay zero. The Toda part contributes nothing to the required long-strip coefficient, completing the proof of Proposition~\ref{prop:ordinary-star-corners}. \subsection{Uniform reddening at the double bond} \label{sec:uniform-double-bond} \begin{theorem}\label{thm:uniform-double-bond} For every pair of positive integers \((a,b)\), the weighted two-strip matrix \(Q(a,b)\) admits a reddening sequence, in either orientation of its double bond. Consequently every weighted spectral tree with one double bond, arbitrary ordinary branches and a single framing admits a reddening sequence. \end{theorem} \begin{proof} We construct reddening words for \(-Q(m,m)\) by induction on \(m\ge2\). For the base, order the vertices as \((s_{0,1},t_{0,1},s_{1,1},t_{1,1},e)\). The matrix is \[ -Q(2,2)=\begin{pmatrix} 0&2&0&0&-2\\ -2&0&0&0&2\\ 0&0&0&2&-1\\ 0&0&-2&0&1\\ 1&-1&1&-1&0 \end{pmatrix}. \] The chronological word \[ (5,4,2,5,4,2,5,1,4,3,5,2,3,1,4,3,1,4,1,2,3,5) \] has all 22 mutations green and terminal coefficient matrix \(-P\), where \(P\) interchanges positions 4 and 5. Direct matrix mutation verifies this assertion. Suppose \(-Q(m,m)\) has a reddening word \(R_m\). The ordinary star of dimensions \((m,m,m)\) has long-strip coefficient two by Proposition~\ref{prop:ordinary-star-corners}(1). The conditional bound \eqref{eq:long-covering-corner} applies because the folded reddening is already supplied by the induction hypothesis. It gives long-strip coefficient at most two for \(R_m\). Lemma~\ref{lem:conditional-strip-extension} therefore constructs a reddening word for \(-Q(m+1,m)\). Now apply the covering bound to this newly obtained negative cluster. Proposition~\ref{prop:ordinary-star-corners}(2) and \eqref{eq:short-covering-corner} give short-strip coefficient at most \(2+0=2\). A short-end application of Lemma~\ref{lem:conditional-strip-extension} constructs a reddening word \(R_{m+1}\) for \(-Q(m+1,m+1)\). This completes the induction. For arbitrary positive \(a,b\), choose \(m\ge\max(2,a,b)\). The explicit long-prefix and short-suffix embedding above identifies \(Q(a,b)\) as a principal submatrix of \(Q(m,m)\). Reddening heredity \cite[Lemma~4.9]{CaoLiTriangular} proves the first assertion. When \(a=b=1\), the matrix is \((0)\), and its sole mutation already makes its vertex red. The isolated reflection of strip indices exchanges the two orientations, and reversing and transporting a reddening word exchanges \(Q\) and \(-Q\). Finally Proposition~\ref{prop:ordinary-branch-completion} attaches every ordinary branch and the framing. \end{proof} Each strip extension changes the word length from \(\ell\) to \(3\ell+2\). The balanced word constructed from the 22-letter base therefore has length \(23\cdot9^{m-2}-1\) for \(m\ge2\). \subsection{Odd orthogonal components} \begin{lemma}\label{lem:odd-B-chamber-reddening} Let \(r>1\) be odd in type \(B_n\). If the neutral residual seed of type \(B_{n-1}\) at \(r-1\) admits reddening, so does the full spectral seed of type \(B_n\) at \(r\). \end{lemma} \begin{proof} Use the chamber sign convention of \cite[sections ``The exchange matrix and boundary minors'' and ``Odd orthogonal components'']{FeiBCFG}. In addition to the residual mutable variables write \(p_2,\ldots,p_n\) for its principal minors and \(A_1,\ldots,A_{n-1}\) for the ascending chambers. These are exactly the additional mutable variables in the full seed. Consider the successively larger principal submatrices obtained by adjoining them in the order \[ p_2,\ldots,p_n,\ A_{n-1},A_{n-2},\ldots,A_1. \] The initial matrix on the residual mutable variables is the negative of the neutral principal matrix, and hence also admits reddening. At each step we shall mutate only at positions already present and make the new vertex a source or a sink relative to them. Mutation invariance and the triangular-extension theorem for skew-symmetrizable matrices then prove the result by induction \cite[Theorem~4.5, Remark~4.6]{CaoLiTriangular}. Fix \(2\le i