Quantum $K$-theoretic divisor axiom for flag manifolds

Basic Information

• Paper ID: 2505.16150
• Title: Quantum $K$-theoretic divisor axiom for flag manifolds
• Authors: Cristian Lenart, Satoshi Naito, Daisuke Sagaki, Weihong Xu, with an appendix by Leonardo C. Mihalcea and Weihong Xu
• Classification: math.QA (Quantum Algebra), math.AG (Algebraic Geometry), math.CO (Combinatorics), math.KT (K-Theory), math.RT (Representation Theory)
• Publication Date: May 2025 (arXiv v2: October 31, 2025)
• Paper Link: https://arxiv.org/abs/2505.16150

Abstract

This paper proves an identity for (torus-equivariant) 3-point, genus-0 $K$-theoretic Gromov-Witten invariants of flag manifolds $G/P$, which can be viewed as a substitute for the "divisor axiom" in (torus-equivariant) quantum $K$-theory. This identity enables the computation of these invariants when two insertions are Schubert classes and another is a Schubert divisor class. The authors' type-independent proof utilizes the Chevalley formula for the (torus-equivariant) quantum $K$-theory ring of flag manifolds, which computes multiplication with Schubert divisor classes via the quantum Bruhat graph.

Research Background and Motivation

Problem Background

1. Limitations of the Classical Divisor Axiom: In cohomological Gromov-Witten theory, there exists the famous divisor axiom: $\langle[\Omega_1], \ldots, [\Omega_{m-1}], [D]\rangle^Y_d = \left(\int_d [D]\right) \cdot \langle[\Omega_1], \ldots, [\Omega_{m-1}]\rangle^Y_d$ where $D$ is a divisor. However, no general analogue of such an axiom exists in the $K$-theory framework.
2. Buch-Mihalcea Conjecture: For type $A$ Lie algebras, Buch and Mihalcea proposed a conjecture for the $K$-theoretic divisor axiom (Conjecture 1.1):\langle[O_{\Omega_1}], [O_{\Omega_2}]\rangle^Y_d & \text{if } d_i > 0, \\ \langle O_{s_i} \cdot [O_{\Omega_1}], [O_{\Omega_2}]\rangle^Y_d & \text{if } d_i = 0, \end{cases}$$ where $O_{s_i} = [O_{Y^{s_i}}]$ is a Schubert divisor.
3. Limitations of Existing Results:
   • The conjecture was originally proposed for type $A$ flag manifolds
   • Counterexamples were found in type $G_2$ (LM, Section 4), showing that the first case does not always hold
   • Previously proven only for cominimal flag manifolds, specific Grassmannians, and symplectic Grassmannians $SG(2,2n)$

Research Motivation

This paper aims to:

1. Provide a type-independent proof: Establish the quantum $K$-theoretic divisor axiom for flag manifolds $G/P$ of all Lie types
2. Characterize correction terms: Give precise correction formulas when the condition $\langle\varpi_i, \theta^\vee\rangle = 1$ is not satisfied
3. Provide computational tools: Utilize the quantum Lakshmibai-Seshadri (QLS) path model to provide computable combinatorial descriptions

Core Contributions

1. Main Theorem (Theorem 3.1): Proves that when $d_i = 0$, the second case of the divisor axiom holds for all $G/P$: $\langle O_{s_i}, O_w, O_x\rangle_d = \langle O_{s_i} \cdot O_w, O_x\rangle_d$
2. Formula with Correction Terms (Theorem 3.2): For general effective degrees $d \in Q^{\vee,+}_K$, gives the precise formula: $\langle O_{s_i}, O_w, O_x\rangle_d = \langle O_w, O_x\rangle_d - \sum_{p \in R^{K;\triangleright}_{w,x,d}} (-1)^{\ell(p)} e^{-\varpi_i + \text{wt}(\eta_p)}$ where the correction term is given through a subset $R^{K;\triangleright}_{w,x,d}$ of QLS paths.
3. Sufficient Conditions: Proves that when $\langle\varpi_i, \theta^\vee\rangle = 1$ and $d_i > 0$, the correction term vanishes, i.e.: $\langle O_{s_i}, O_w, O_x\rangle_d = \langle O_w, O_x\rangle_d$ This covers all minimal and cominimal fundamental weights, as well as all fundamental weights in type $C$.
4. Complete Classification: Provides a complete list of fundamental weights satisfying the condition $\langle\varpi_i, \theta^\vee\rangle = 1$ (Section 2.4).
5. Peterson Comparison Formula (Appendix A, Proposition 1.3): Proves the formula for reducing KGW invariants of $G/P$ to those of $G/B$: $\langle\gamma_1, \ldots, \gamma_m\rangle^{G/P}_d = \langle\pi^*\gamma_1, \ldots, \pi^*\gamma_m\rangle^{G/B}_{\hat{d}}$
6. Positivity Result (Corollary 3.5): In the non-equivariant case, proves positivity properties of KGW invariants $\langle O_{s_i}, O_u, (O^w)^\vee\rangle_d$.

Detailed Methodology

Core Strategy

The paper employs a combinatorial-algebraic approach, with the main technical route being:

1. Structure of the Quantum $K$-theory Ring: Utilizes the relationship between multiplication $\star$ in the quantum $K$-theory ring $QK_T(Y)$ and 3-point KGW invariants: $((σ_1 \star σ_2, σ_3)) = \sum_{d \in Q^{\vee,+}_K} Q^d \langle σ_1, σ_2, σ_3\rangle_d$
2. Chevalley Formula: Crucially utilizes the quantum $K$-Chevalley formula (Theorem 2.17) proven in NOS and LNS: $O_{s_i} \star O_w = O_w + \sum_{v \in W} \sum_{\substack{\eta \in \text{QLS}(\varpi_i) \\ \kappa(\eta,v) = w}} (-1)^{\ell(v) - \ell(w) + 1} e^{-\varpi_i + \text{wt}(\eta)} Q^{\zeta(\eta,v)} O_v$
3. Quantum Bruhat Graph: Encodes combinatorial information of Schubert calculus through the quantum Bruhat graph $\text{QBG}(W)$.

Quantum Lakshmibai-Seshadri Paths

Definition (Definition 2.9): A QLS path of shape $\lambda$ is a pair $\eta = (v; a) = (v_1, \ldots, v_s; a_0, a_1, \ldots, a_s), \quad s \geq 1$ where:

• $v_1, \ldots, v_s \in W^J$ ($J = J_\lambda = \{j \in I \mid \langle\lambda, \alpha^\vee_j\rangle = 0\}$)
• $0 = a_0 < a_1 < \cdots < a_s = 1$ is an increasing sequence of rational numbers
• For each $k = 1, \ldots, s-1$, there exists a directed path from $v_{k+1}$ to $v_k$ in $\text{QBG}^{a_k\lambda}(W^J)$

Key Properties:

• When $\langle\varpi_i, \theta^\vee\rangle = 1$, $\text{QLS}(\varpi_i) = \text{LS}(\varpi_i)$ (contains only ordinary LS paths)
• Weight definition: $\text{wt}(\eta) = \sum_{k=1}^s (a_k - a_{k-1}) v_k \lambda$

Technical Innovations

1. Sijection Method: Constructs two key sign-reversing bijections (sijections) $\Theta$ and $\Theta'$:
   • $\Theta$ acts by adding/removing the last edge labeled $\alpha_i$
   • $\Theta'$ adds/removes edges based on the reflection order relationship between the final label $\beta$ and initial label $\gamma$
2. Stratified Treatment: Decomposes the QLS path set as: $\text{QLS}^{K;\triangleright}_{w,x,d} = \text{QLS}^{K;\triangleright}_{w,x,d,+} \sqcup \text{QLS}^{K;\triangleright}_{w,x,d,0}$ and separately proves:
   • $\sum_{p \in \text{QLS}^{K;\triangleright}_{w,x,d,+}} (-1)^{\ell(p)} e^{-\varpi_i + \text{wt}(\eta_p)} = 0$ (via sijections $\Theta$ and $\Theta'$)
   • $\sum_{p \in \text{QLS}^{K;\triangleright}_{w,x,d,0} \setminus R^{K;\triangleright}_{w,x,d}} (-1)^{\ell(p)} e^{-\varpi_i + \text{wt}(\eta_p)} = 0$ (via sijection $\Psi$)
3. Curve Neighborhoods: Introduces the geometric object $\Gamma_d(Y_u, Y^{s_i})$ (Definition 1.4): $\Gamma_d(Y_u, Y^{s_i}) := \text{ev}_3(M_d(Y_u, Y^{s_i}))$ and proves (Lemma 2.14):\text{ev}_2(\text{ev}_1^{-1}(Y_u)) & \text{if } d_i > 0, \\ \text{ev}_2(\text{ev}_1^{-1}(Y_u)) \cap Y^{s_i} & \text{if } d_i = 0 \end{cases}$$

Proof Outline of Main Theorems

Proof of Theorem 3.1 (Case $d_i = 0$)

Proof Strategy (Section 4.2):

1. Algebraic Expansion: Starting from the quantum metric relation: $\sum_{d \in Q^{\vee,+}_K} Q^d \langle O_{s_i}, O_w, O_x\rangle_d = ((O_{s_i} \star O_w, O_x))$
2. Chevalley Formula Application: Using Corollary 2.19: $O_{s_i} \star O_w = O_w - \sum_{p \in \text{QLS}^\triangleright_w} (-1)^{\ell(p)} e^{-\varpi_i + \text{wt}(\eta_p)} Q^{[\text{qwt}(p)]} O_{\lfloor\text{end}(p)\rfloor}$
3. Key Decomposition: Define the set $U := \{z \in W^{I\setminus K} \mid \lfloor\text{end}(p)\rfloor = z \text{ for some } p \in \text{QLS}^\triangleright_w \text{ with qwt}(p) = 0\}$
4. Modular Arithmetic Analysis: Prove:
   • If $z \in U$, then $a_z(Q) \in c_z + Q_i R(T)[Q]$
   • If $z \notin U$, then $a_z(Q) \in Q_i R(T)[Q]$, hence $c_z = 0$
5. Coefficient Extraction: Since $d_i = 0$ means $Q^d \notin Q_i R(T)[Q]$, extracting the coefficient of $Q^d$ gives: $\langle O_{s_i}, O_w, O_x\rangle_d = \sum_{z \in U} c_z \langle O^z, O_x\rangle_d = \sum_{z \in W^{I\setminus K}} c_z \langle O^z, O_x\rangle_d = \langle O_{s_i} \cdot O_w, O_x\rangle_d$

Proof of Theorem 3.2 (General Case)

Proof Strategy (Section 4.3):

1. Fundamental Identity (Equations 4.10-4.11): Via the Chevalley formula and 2-point invariant formula (Lemma 4.1): $\langle O_{s_i}, O_w, O_x\rangle_d = \langle O_w, O_x\rangle_d - \sum_{p \in \text{QLS}^{K;\triangleright}_{w,x,d}} (-1)^{\ell(p)} e^{-\varpi_i + \text{wt}(\eta_p)}$
2. Set Decomposition: $\text{QLS}^{K;\triangleright}_{w,x,d,+} := \{p \in \text{QLS}^{K;\triangleright}_{w,x,d} \mid \langle\varpi_i, d - \text{qwt}_2(p)\rangle > 0\}$$\text{QLS}^{K;\triangleright}_{w,x,d,0} := \{p \in \text{QLS}^{K;\triangleright}_{w,x,d} \mid \langle\varpi_i, d - \text{qwt}_2(p)\rangle = 0\}$
3. Proving the Contribution of $\text{QLS}^{K;\triangleright}_{w,x,d,+}$ Vanishes (Section 4.3.1):
   • Construct sijection $\Theta$: by adding/removing the last edge labeled $\alpha_i$
   • Partition the set into $A$ (last label is $\alpha_i$) and $B$ (last label is not $\alpha_i$)
   • Further decompose $B = B_1 \sqcup B_2$
   • Apply $\Theta$ to $A \sqcup B_1$, construct new sijection $\Theta'$ for $B_2$
4. Proving the Contribution of $\text{QLS}^{K;\triangleright}_{w,x,d,0} \setminus R^{K;\triangleright}_{w,x,d}$ Vanishes (Section 4.3.2):
   • Construct sijection $\Psi$: based on the reflection order relationship between the final label $\beta$ and the initial label $\gamma$ of the path to $x$
   • Key observation: $(β, γ) \neq (\alpha_i, \alpha_i)$ (otherwise leads to contradiction)
5. Sufficient Conditions for Correction Term Vanishing (Section 4.3.3):
   • When $\langle\varpi_i, \theta^\vee\rangle = 1$, $\text{QLS}(\varpi_i) = \text{LS}(\varpi_i)$
   • Can take $N = 2$, and $\eta_p = (\lfloor\text{end}(p_2)\rfloor^J, \lfloor w\rfloor^J)$ satisfies $\lfloor\text{end}(p_2)\rfloor^J \geq \lfloor w\rfloor^J$
   • By Lemma 2.6, get $\text{end}(p_2) \geq w$, hence $\text{qwt}_2(p) = 0$
   • Thus $\langle\varpi_i, d - [\text{qwt}_2(p)]\rangle = d_i > 0$, i.e., $R^{K;\triangleright}_{w,x,d} = \emptyset$

Key Technical Details

Quantum Bruhat Graph

Definition 2.1: For a subset $L \subseteq I$, the quantum Bruhat graph $\text{QBG}(W^L)$ is a $(\Delta^+ \setminus \Delta^+_L)$-labeled directed graph with:

• Vertices: elements of $W^L$
• Edges: $x \xrightarrow{\alpha} y$ ($x, y \in W^L$, $\alpha \in \Delta^+ \setminus \Delta^+_L$), satisfying $y = \lfloor xs_\alpha\rfloor^L$ and:
  • (B) Bruhat edge: $\ell(y) = \ell(x) + 1$
  • (Q) Quantum edge: $\ell(y) = \ell(x) + 1 - 2\langle\rho - \rho_L, \alpha^\vee\rangle$

Key Properties (Theorem 2.3):

• There exist unique label-increasing and label-decreasing shortest paths
• Quantum weight $\text{qwt}(v \Rightarrow w)$ is well-defined (independent of shortest path choice)

Double-Tilted Bruhat Order

Definition 2.4: For each $v \in W$, define the double $v$-tilted Bruhat order $\leq^*_v$: $w_1 \leq^*_v w_2 \Longleftrightarrow \ell(w_1 \Rightarrow v) = \ell(w_1 \Rightarrow w_2) + \ell(w_2 \Rightarrow v)$

Important Result (Proposition 2.5): Each coset $uW_L$ has a unique $\leq^*_v$-maximal element, denoted $\max(uW_L, \leq^*_v)$.

Sijection Construction Details

Sijection $\Theta$ (for $A \sqcup B_1$):

For $p = (p_N, \ldots, p_2, p_1) \in \text{QLS}^\triangleright_w$:

1. If $\kappa_L(p_1) = \alpha_i$: remove the last edge of $p_1$ (labeled $\alpha_i$)
2. If $\kappa_L(p_1) \neq \alpha_i$: add an edge labeled $\alpha_i$ at the end of $p_1$

Sijection $\Theta'$ (for $B_2$):

Set $p \in B_2$, define:

• $\beta := \kappa_L(p_1)$ (final label of $p_1$, or $-\infty$ if $p_1$ is trivial)
• $\gamma := \iota_L(\text{dec}(z \Rightarrow x))$ (initial label of the label-decreasing path from $z = \text{end}(p)$ to $x$, or $-\infty$ if $z = x$)

Key observation: $\beta \neq \gamma$ (Remark 4.5)

1. If $\beta \triangleright \gamma$: remove the last edge of $p_1$ (labeled $\beta$)
2. If $\beta \triangleleft \gamma$: add an edge labeled $\gamma$ at the end of $p_1$

Verification: $\Theta'(p) \in B_2$ (key is checking conditions 4.21 and 4.22)

Applications and Corollaries

Positivity Result

Corollary 3.5: If $d_i = 0$ or $\langle\varpi_i, \theta^\vee\rangle = 1$, then the non-equivariant KGW invariant satisfies: $(-1)^{\ell(w) - \text{codim} \Gamma_d(Y_u, Y^{s_i})} \langle O_{s_i}, O_u, (O^w)^\vee\rangle_d \geq 0$

Proof Outline:

1. By Theorems 3.1 and 3.2, reduce to computing $\chi_Y(O_{\Gamma_d(Y_u, Y^{s_i})} \cdot (O^w)^\vee)$
2. Apply Brion's theorem Br: the expansion of the structure sheaf of a closed irreducible subvariety with rational singularities in the Schubert basis has alternating signs
3. For the equivariant case, use the generalization by Anderson-Griffeth-Miller AGM

Peterson Comparison Formula

Proposition 1.3: Let $\pi: G/B \to G/P$ be the natural projection, and $\hat{d}$ be the Peterson lift of $d$, then: $\langle\gamma_1, \ldots, \gamma_m\rangle^{G/P}_d = \langle\pi^*\gamma_1, \ldots, \pi^*\gamma_m\rangle^{G/B}_{\hat{d}}$

Proof Highlights (Appendix A):

1. Use Woodward's result W: the natural map $M_{0,m}(G/B, \hat{d}) \to M_{0,m}(G/P, d)$ is cohomologically trivial (Corollary A.4)
2. Apply the projection formula and pushforward of structure sheaves

Corollary A.6: Theorems 3.1 and 3.2 hold for $G/B$ $\Rightarrow$ they hold for $G/P$

Concrete Examples

Example 3.3 (Counterexample in Type $G_2$)

Consider $g$ of type $G_2$, $i = 2$ (long simple root), $w = s_2s_1s_2s_1s_2$, $d = d_1\alpha^\vee_1 + 2\alpha^\vee_2$ ($d_1 > 0$).

Computation:

• $\langle\varpi_2, \theta^\vee\rangle = 2 \neq 1$ ($\theta = 3\alpha_1 + 2\alpha_2$)
• $N_2 = 6$, $\text{QLS}(\varpi_2) \neq \text{LS}(\varpi_2)$

Result: