Matsuki duality for loop groups

Technical Summary: Matsuki Duality for Loop Groups

Problem Statement
The paper addresses the extension of the classical Matsuki duality to the setting of loop groups. Classically, for a complex connected reductive group $G$ with a real form $G_{\mathbb{R}}$ and a symmetric subgroup $K$, Matsuki duality establishes a bijection between $K$-orbits and $G_{\mathbb{R}}$-orbits on the flag variety $B$. The authors seek to establish an analogous bijection for loop groups, specifically relating orbits of the loop group of the symmetric subgroup, $K((t))$, and the polynomial loop group of the real form, $L G_{\mathbb{R}}$, on the affine Grassmannian ($Gr$) and the affine flag variety ($F\ell$).

A central challenge in this setting is the presence of "twisted" involutions. The authors consider two involutions on the loop group $G(F)$ (where $F = \mathbb{C}((t))$):

1. $\theta$: A twisted involution defined by $\theta(\gamma)(t) = \theta_0(\gamma(\epsilon t))$, where $\theta_0$ is a complex Cartan involution and $\epsilon \in \{1, -1\}$.
2. $\eta$: A twisted involution on the polynomial loop group $G[t, t^{-1}]$ defined by $\eta(\gamma)(t) = \eta_0(\gamma(\epsilon \bar{t}^{-1}))$, where $\eta_0$ is the conjugation defining the real form $G_{\mathbb{R}}$.

The goal is to classify the orbits of $G(O)$ (formal arc group) and $I$ (Iwahori subgroup) under $\theta$-twisted conjugation on the anti-fixed point set $G(F)^{\text{inv} \circ \theta}$, and to relate them to the orbits of $G[t]$ and $I_p$ (polynomial Iwahori) under $\eta$-twisted conjugation on $G[t, t^{-1}]^{\text{inv} \circ \eta}$.

Methodology
The approach is entirely combinatorial, inspired by Nadler's work on affine Grassmannians and the authors' previous work on singularities of orbit closures. The methodology proceeds through the following steps:

1. Cartan Decomposition and Parametrization: The authors utilize the Cartan decomposition to stratify the anti-fixed point sets into double cosets indexed by dominant coweights $\lambda \in X^*(T)^+$.
2. Orbit Classification: For each stratum, they reduce the classification of orbits to the study of specific subspaces ($A_{\lambda,0}$ and $B_{\lambda,0}$) within Levi subgroups $L_\lambda$. They prove that the orbits are in bijection with the connected components of these subspaces, which are further reduced to orbits under compact real forms ($L_{\lambda, c}$ or $T_c$).
3. Intersection Analysis: A crucial technical component is the analysis of the intersection between a $\theta$-twisted orbit and an $\eta$-twisted orbit. The authors demonstrate that the intersection of corresponding orbits is a single orbit of the compact group $G_c$ (for spherical orbits) or $T_c$ (for Iwahori orbits).
4. Uniformization and Twistor Geometry: The paper connects these orbit classifications to the geometry of vector bundles on real forms of the projective line $\mathbb{P}^1$. By considering the conjugation $\eta$ on $\mathbb{P}^1$, the authors identify the set of orbits with isomorphism classes of $G_{\mathbb{R}}$-bundles on the real form $\mathbb{P}^1_\eta$ (which is either the real projective line or the "twistor-$\mathbb{P}^1$" depending on $\epsilon$).
5. Kottwitz Sets: The results are linked to the Kottwitz set $B(G_{\mathbb{R}})$, providing a geometric interpretation of these sets in terms of loop group orbits.

Key Contributions and Results

• Main Theorem (Matsuki Duality for Loop Groups): The paper establishes a bijection between:

  • $G(O)$-orbits on $G(F)^{\text{inv} \circ \theta}$ and $G[t]$-orbits on $G[t, t^{-1}]^{\text{inv} \circ \eta}$.
  • $I$-orbits on $G(F)^{\text{inv} \circ \theta}$ and $I_p$-orbits on $G[t, t^{-1}]^{\text{inv} \circ \eta}$.
    The correspondence is characterized by the property that the intersection of corresponding orbits is a single orbit of the compact group $G_c$ (or $T_c$). Furthermore, there is a bijection between equivariant local systems on corresponding orbits.

• Affine Grassmannian and Flag Varieties: By restricting to the orbit through the base point, the authors derive a Matsuki duality for the affine Grassmannian ($Gr$) and affine flag variety ($F\ell$). This generalizes Nadler's result for the un-twisted case ($\epsilon = 1$) to the twisted case ($\epsilon = -1$) and extends the duality to affine flag varieties, a setting not previously covered even for $\epsilon = 1$.

• Vector Bundles and Twistor Geometry: The authors provide a uniformization theorem identifying the orbit sets with:

  • Isomorphism classes of $G_{\mathbb{R}}$-bundles on the real form $\mathbb{P}^1_\eta$.
  • Isomorphism classes of such bundles equipped with Borel reductions at specific points.
    This connects the combinatorial orbit data to the geometry of "circular" or "antipodal" real twistor structures (in the sense of Simpson).

• Kottwitz Sets: The paper identifies the set of $G[t]$-orbits with the Kottwitz set $B(G_{\mathbb{R}})$, offering a new geometric perspective on these sets via loop group orbits.

• Examples: The theory is explicitly worked out for the real forms of $GL_2$, including $GL_2(\mathbb{R})$, $GL_1(\mathbb{H})$, $U(2)$, and $U(1, 1)$. These examples illustrate the parametrization of orbits and the structure of the corresponding vector bundles, including cases where the parametrizing sets are empty or non-trivial.

Significance and Claims
The authors claim that this work provides a foundational combinatorial framework for understanding the geometry of real loop groups. The significance lies in:

1. Unification: It treats all real forms within the same pure inner class simultaneously.
2. New Geometric Connections: It establishes a direct link between loop group orbit intersections and vector bundles on twistor spaces, bridging representation theory and algebraic geometry.
3. Foundation for Derived Equivalences: The paper notes that these results serve as a precursor to a forthcoming work ([CNY]) which will use Morse-theoretic interpretations to upgrade this duality to an equivalence of equivariant derived categories. This is presented as a key ingredient for relating geometric Langlands for real groups to relative Langlands duality.

The paper does not claim to solve the full geometric Langlands conjecture for real groups but positions its combinatorial results as a necessary step toward such a program, specifically by clarifying the orbit structures that underpin the derived equivalences.