🔢 mathematics

Emerton--Gee stacks, Serre weights, and Breuil--Mézard conjectures for $\mathrm{GSp}_4$

This paper constructs a moduli stack of rank 4 symplectic $(\varphi,\Gamma)$ -modules and, by analyzing its geometry via local models, proves the $\mathrm{GSp}_4$ -analogue of the Breuil–Mézard conjecture, the weight part of Serre's conjecture, and a modularity lifting result for tamely potentially crystalline representations under genericity conditions.

Original authors: Heejong Lee

Published 2026-07-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Heejong Lee

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, invisible tapestry woven from numbers and shapes. In the world of mathematics, there is a famous puzzle called the "Modularity Conjecture." It suggests that every strange, jagged pattern you can draw using numbers (specifically, Galois representations, which are like secret codes describing how numbers twist and turn) is actually just a shadow cast by a smooth, rhythmic wave (a modular form). For a long time, mathematicians could only see these shadows clearly when the patterns were simple, like drawing on a flat sheet of paper. But what happens when the paper gets crumpled, or when the drawing gets incredibly complex?

To solve this, mathematicians use a powerful tool called the "Langlands Program." Think of this as a universal translator that tries to speak two different languages: the language of symmetry (groups like $GSp_4$ , which describe how four-dimensional objects can be stretched and twisted while keeping their shape) and the language of number theory. A crucial part of this translation is figuring out the "weight" of a pattern. In everyday terms, "weight" isn't about how heavy something is; it's like the color or the texture of the pattern. If you know the weight, you know exactly which smooth wave created the shadow. For simple shapes, we have a perfect dictionary for this. But for complex, four-dimensional shapes, the dictionary was missing pages. This paper steps in to write those missing pages, specifically for a group of symmetries called $GSp_4$ , which governs a specific type of four-dimensional dance.

The Paper: A Map for a Four-Dimensional Dance Floor

In this paper, Heejong Lee builds a massive, digital map of a mathematical landscape that no one had fully explored before. Imagine a giant, multi-layered dance floor where dancers (representations of numbers) move in four dimensions. Some dancers move smoothly and predictably (crystalline), while others are a bit more chaotic. Mathematicians want to know: "If I see a dancer moving in a specific, messy way, can I predict exactly what their smooth, perfect version looks like?"

To do this, Lee constructs a new kind of "Emerton–Gee stack." If a stack is a pile of data, this one is a magical, infinite library where every book represents a possible way these four-dimensional numbers can behave. The author proves that this library is well-organized (it's a "Noetherian formal algebraic stack") and that every single book in it can be traced back to a specific "Serre weight" (the color/texture of the pattern). This is like proving that every possible dance move in the library has a unique ID card.

The Local Models: The Blueprint
The real magic happens when Lee looks at the "local models." Imagine you are trying to understand a complex building, but you can only see the outside. Lee builds a transparent, geometric blueprint (a "local model") that lets you see the inside structure of these mathematical buildings. He focuses on the "torus fixed points," which are like the corners of the building where the structure is most rigid.

He proves a surprising fact: at these corners, the building is "unibranch." In plain English, this means that if you stand at a corner and look at the floor, you don't see a fork in the road where the path splits into two different directions. There is only one path. This might sound boring, but in the world of these equations, it's a huge deal. It proves that the "deformation rings" (the mathematical tools used to stretch and change the numbers) are "domains." Think of a domain as a solid, unbroken piece of land. If the land is broken (not a domain), you can't build a stable house on it. Lee shows that under certain "generic" conditions (basically, when the numbers aren't weirdly special or broken), the land is solid and unbroken.

The Big Wins: Three New Bridges
With this solid ground, Lee builds three major bridges to connect different islands of mathematics:

The Breuil–Mézard Conjecture: This is a rule that predicts how complex a mathematical object is based on its "mod p" (remainder) version. Lee proves this rule works for these four-dimensional shapes. It's like saying, "If you know the rough sketch of a sculpture, you can now accurately predict how much marble was needed to carve the final masterpiece."
The Weight Part of Serre's Conjecture: This is the "dictionary" mentioned earlier. Lee proves that for these four-dimensional shapes, the "weight" (the color/texture) is determined entirely by how the shape behaves at a specific point (the inertia group). He confirms that the dictionary entries match the patterns perfectly, provided the numbers aren't too messy.
Modularity Lifting: This is the ultimate goal. If you find a pattern that looks like it might be a shadow of a smooth wave, Lee's work proves that if it passes a few specific tests, it is a shadow. It's like finding a footprint in the mud and proving, beyond a doubt, that it belongs to a specific famous explorer.

The Catch (The "Generic" Condition)
It's important to note that these results come with a condition. Lee says these proofs work when the numbers are "sufficiently generic." In our analogy, this means the dancers are doing a standard, well-practiced routine. If the dancers are doing something bizarrely weird or broken (non-generic), the map might not hold up. The paper doesn't claim to solve the problem for every possible weird case, but it solves it for the vast majority of "normal" cases, which is a massive step forward.

Why It Matters
This isn't just about drawing pretty shapes. By proving that these four-dimensional mathematical structures are solid (domains) and that we can predict their weights, Lee gives mathematicians the tools to solve even bigger problems. It's like giving an architect a blueprint that proves a skyscraper won't collapse, so they can finally start building the next floor. This work connects deep theories about numbers, shapes, and symmetries, bringing us closer to understanding the fundamental rhythm of the mathematical universe.

Technical Summary: Emerton–Gee Stacks, Serre Weights, and Breuil–Mézard Conjectures for GSp4

1. Problem Statement and Context
The paper addresses the extension of the mod $p$ Langlands program and the weight part of Serre's conjecture from the general linear group $GL_n$ to the symplectic similitude group $GSp_4$ . While the theory of Emerton–Gee stacks (moduli stacks of $p$ -adic Langlands parameters) and local models for potentially crystalline deformation rings has been established for $GL_n$ (notably by Le, Le Hung, Levin, and Morra), the case for $GSp_4$ presents significant structural challenges. Specifically, $GSp_4$ possesses a non-abelian unipotent radical in its Klingen parabolic subgroup, complicating the construction of families of extensions and the analysis of deformation rings. The primary objective is to construct the Emerton–Gee stack for $GSp_4$ , develop a theory of local models for its potentially crystalline substacks, and apply these geometric tools to prove the Breuil–Mézard conjecture and the weight part of Serre's conjecture for global Galois representations valued in $GSp_4$ .

2. Methodology
The author employs a combination of $p$ -adic Hodge theory, algebraic geometry, and global patching arguments:

Construction of Symplectic Emerton–Gee Stacks: The paper defines a moduli stack $X_{Sym, K}$ of rank 4 symplectic projective étale $(\phi, \Gamma)$ -modules. This stack algebraically interpolates continuous representations $\rho: G_K \to GSp_4(A)$ . The construction relies on generalizing the results of Emerton and Gee for $GL_4 \times GL_1$ to the symplectic setting, utilizing the equivalence between symplectic modules and $GSp_4$ -valued representations.
Local Models and Geometry: To study the geometry of potentially crystalline substacks, the author adapts the theory of local models developed for $GL_n$ . These models are constructed as closed subvarieties within Pappas–Zhu local models associated with the group $Res_{K/\mathbb{Q}_p} GSp_4$ . A critical step involves proving the unibranch property at torus-fixed points of these local models. This requires analyzing the special fiber of naive local models in mixed characteristic and matching irreducible components with Serre weights via combinatorial data involving the extended affine Weyl group.
Inertial Local Langlands Correspondence: A key technical component is the construction of an inertial local Langlands correspondence for $GSp_4$ . The author associates a tame type $\sigma(\tau)$ (an irreducible representation of $GSp_4(k)$ ) to a tame inertial $L$ -parameter $\tau$ . This involves explicit descriptions of Deligne–Lusztig representations and their behavior under parabolic induction (Siegel and Klingen parabolics) and theta correspondences.
Global Patching and Lifting: The paper utilizes a patching argument (following Taylor–Wiles and Calegari–Geraghty) to construct weak patching functors. A crucial ingredient is the proof of the existence of crystalline lifts for continuous representations $G_K \to GSp_4(\mathbb{F})$ . This relies on obstruction theory developed by Lin for reductive groups, specifically handling the non-abelian cohomology issues arising from the Klingen parabolic.

3. Key Contributions and Results

Geometric Properties of $X_{Sym, K}$ :
- The stack $X_{Sym, K}$ is proven to be a Noetherian formal algebraic stack.
- Closed substacks $X^{\lambda, \tau}_{Sym, K}$ parameterizing potentially crystalline representations with Hodge type $\lambda$ and inertial type $\tau$ are constructed and shown to be $p$ -adic formal algebraic stacks flat over $\mathcal{O}$ .
- The underlying reduced substack $X_{Sym, K, red}$ is shown to be equidimensional of dimension $4[K:\mathbb{Q}_p]$ , with irreducible components naturally labeled by Serre weights.
Local Model Theory and Unibranch Property:
- For a regular Hodge type $\lambda$ and a sufficiently generic tame inertial type $\tau$ , the local model $M(\lambda, \nabla^a_\tau)$ is proven to be unibranch at any torus-fixed point.
- Corollary: This geometric result implies that the potentially crystalline deformation ring $R^{\lambda, \tau}_\rho$ is a domain (if non-zero) for tame $\rho$ . This is a pivotal result for the patching argument, ensuring that the support of patched modules is the entire spectrum of the deformation ring.
Inertial Local Langlands for $GSp_4$ :
- Theorem 2.4.1 establishes that if a smooth representation $\pi$ of $GSp_4(K)$ contains a specific tame type $\sigma(\tau)$ , then the restriction of its $L$ -parameter to inertia is isomorphic to $\tau$ . This result is essential for linking local deformation rings to global automorphic forms.
Applications to Conjectures:
- Breuil–Mézard Conjecture: The geometric Breuil–Mézard conjecture is formulated and proven for tamely potentially crystalline deformation rings under genericity conditions. The multiplicity of a Serre weight $\sigma$ in the special fiber of the deformation ring is expressed as a cycle intersection number involving the local model geometry.
- Weight Part of Serre's Conjecture: For global Galois representations $r: G_F \to GSp_4(\mathbb{F})$ satisfying Taylor–Wiles conditions and tame genericity at $p$ , the set of modular Serre weights $W(r)$ is proven to equal the combinatorially defined set $W^?(\rho|_{I_F})$ .
- Modularity Lifting: A modularity lifting theorem is proven for tamely potentially crystalline representations with regular Hodge type and sufficiently generic inertial types. If a representation $r'$ is automorphic and congruent to $r$ modulo $p$ , then $r$ is also automorphic.

4. Significance and Claims
The paper claims to generalize the foundational framework of Emerton–Gee stacks and local models from $GL_n$ to $GSp_4$ , a step necessary for advancing the mod $p$ Langlands program for symplectic groups. The significance lies in:

Overcoming Non-Abelian Obstacles: Successfully navigating the non-abelian unipotent radical of the Klingen parabolic to construct families of extensions and prove the existence of crystalline lifts, which was previously an open difficulty for $GSp_4$ .
Geometric Rigor: Establishing the unibranch property of local models for $GSp_4$ , which is a non-trivial geometric property required to deduce that deformation rings are domains. This allows the application of patching techniques that were previously restricted to $GL_n$ .
Verification of Conjectures: Providing the first proofs of the Breuil–Mézard conjecture and the weight part of Serre's conjecture for $GSp_4$ under genericity assumptions, thereby extending the scope of these deep number-theoretic results beyond the general linear group.

The work relies heavily on the "genericity" of the inertial types and Hodge weights relative to $p$ , a condition explicitly stated as necessary for the geometric and combinatorial arguments to hold. The results are presented as conditional on the existence of certain global automorphic lifts and the validity of the Taylor–Wiles conditions, consistent with the current state of the art in modularity lifting.

More like this