🔢 mathematics

Chow rings, cohomology rings, and point counts of moduli spaces of curves

This expository article surveys state-of-the-art results concerning the Chow rings, cohomology rings, and finite field point counts of the moduli spaces of smooth and stable curves, specifically examining the conditions under which these invariants are tautological or polynomial in the field size.

Original authors: Hannah Larson

Published 2026-06-30

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Hannah Larson

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 you are an architect trying to understand a vast, shifting city called the Moduli Space of Curves. This isn't a city made of brick and mortar, but a mathematical "city" where every single building represents a unique shape of a smooth, curved loop (like a rubber band) with some number of dots (marked points) attached to it.

The author, Hannah Larson, is writing a guidebook to this city. She wants to answer three big questions about the city's structure, using three different "lenses" or tools to look at it:

The Chow Ring (The Blueprint): This looks at the city's algebraic structure. It asks: "If I take two specific shapes and glue them together, what new shape do I get? How do these shapes intersect?" It's like counting the bricks and beams in the buildings.
The Cohomology Ring (The Topology): This looks at the city's shape and holes. It asks: "Does this city have tunnels? Are there loops you can't shrink to a point?" It's like studying the geography and the "holes" in the landscape.
Point Counts (The Census): This looks at the city over a finite field (a world with a limited number of elements, like a video game with a fixed number of pixels). It asks: "If we count the number of valid shapes in this limited world, does the count follow a simple, predictable pattern (a polynomial)?"

The "Tautological" Neighborhood

The paper focuses heavily on a special district in this city called the Tautological Ring.

Think of the city as having a "Main Street" built from a few standard, easy-to-understand blocks. These blocks are created by simple operations:

Gluing: Taking two curves and sticking them together at marked points.
Forgetting: Taking a curve with a dot and just erasing the dot.

The Tautological Ring is the collection of all shapes you can build using only these standard blocks and operations. The big question the paper asks is: Is the entire city just made of these standard blocks, or are there weird, exotic buildings that can't be built from the standard kit?

The Three Main Findings

1. When is the city "Standard"? (Chow and Cohomology)

The paper maps out exactly which versions of this city (defined by the number of loops $g$ and dots $n$ ) are made entirely of standard blocks.

Small Cities are Simple: For small numbers of loops and dots (like a city with 1 loop and 10 dots), the entire city is "tautological." Every shape can be built from the standard blocks.
Big Cities get Weird: As the city gets bigger (more loops or dots), strange, non-standard buildings start appearing.
- The "Odd" Anomaly: The paper highlights a specific weirdness in a city with 1 loop and 11 dots. It has a "hole" (an odd cohomology group) that cannot be built from standard blocks. This is like finding a hidden tunnel in a city that was supposed to be flat.
- The "Double Cover" Anomaly: In larger cities, there are shapes formed by "double covers" (like a map that folds over itself twice) that are also non-standard.

The paper provides a massive chart (a grid of $g$ and $n$ ) showing exactly where the city is "all standard" (filled circles) and where it has "weird buildings" (red X's).

2. The Connection Between Blueprints and Geography

The paper explains a deep link between the Blueprint (Chow) and the Geography (Cohomology).

The Ideal Scenario: If the city is "proper" (a closed, compact city with no open edges) and the blueprint is simple (all standard), then the geography is also simple. The "holes" in the city correspond perfectly to the standard blocks.
The Reality Check: If the city is open (like the interior of the city, $M_{g,n}$ , which has edges), things get messy. You can have a simple blueprint but a complex geography with hidden tunnels (odd cohomology) that the blueprint doesn't show. The paper shows that for many large cities, the geography is not just a reflection of the standard blocks.

3. The Census and the "Polynomial" Pattern

The third lens is counting the points.

The Polynomial Rule: In many mathematical cities, if you count the number of shapes for different sizes of the "finite world" ( $q$ ), the answer is a simple polynomial (like $q^2 + 2q + 1$ ). This happens when the city's geography is "Tate type" (very regular, no weird tunnels).
The Breakdown: The paper finds that for small cities, the count is a perfect polynomial. But once you hit certain thresholds (like 1 loop with 11 dots), the count stops being a simple polynomial. It starts having "wobbles" or corrections that look like complex musical notes (related to modular forms).
The Conjecture: The authors propose a rule: The city has a simple, polynomial census count if and only if the city is small enough (specifically, if $3g + 2n < 25$ ). Once the city gets too big, the census becomes too complex to be a simple polynomial.

The "Gluing" Strategy

How did they figure this out? They used a strategy like building with LEGO.

Base Cases: They first solved the problem for the smallest, simplest cities (low genus, few dots).
Induction: They showed that if you know the rules for small cities, you can figure out the rules for bigger cities by gluing the small ones together.
The Boundary: They realized that the "weird" parts of a big city often come from the "boundary" (where curves break apart or touch). By understanding the boundary, they could understand the whole city.

Summary

In short, this paper is a map of a mathematical universe. It tells us:

Where the rules are simple: For small numbers of loops and dots, everything is predictable and built from standard parts.
Where the rules break: As the numbers grow, "exotic" shapes and hidden tunnels appear that defy simple construction.
The limit of predictability: There is a specific size limit ( $3g + 2n < 25$ ) beyond which the simple counting formulas stop working, replaced by complex, non-polynomial patterns.

The paper doesn't tell us how to build a bridge or cure a disease; it tells us the fundamental limits of predictability in the geometry of curved shapes. It draws a line in the sand between the "simple, orderly world" of small moduli spaces and the "complex, chaotic world" of large ones.

Technical Summary: Chow Rings, Cohomology Rings, and Point Counts of Moduli Spaces of Curves

Problem Statement
This expository article investigates three closely related invariants of the moduli spaces of curves: the Chow ring $A^*(M)$ , the cohomology ring $H^*(M)$ , and the point counts over finite fields $\#M(\mathbb{F}_q)$ . The primary objects of study are $M_{g,n}$ , the moduli space of smooth genus $g$ curves with $n$ marked points, and $\overline{M}_{g,n}$ , its Deligne–Mumford compactification by stable curves. The central problem is to determine the extent to which these invariants are "tautological." Specifically, the paper addresses three questions:

For which $(g, n)$ is the Chow ring (or cohomology ring) generated by tautological classes?
For which $(g, n)$ is the point count $\#M_{g,n}(\mathbb{F}_q)$ (or $\#\overline{M}_{g,n}(\mathbb{F}_q)$ ) given by a polynomial in $q$ ?
How do these invariants relate to one another, particularly in the presence of non-tautological classes or non-polynomial point counts?

Methodology and Framework
The paper synthesizes recent advances in algebraic geometry, arithmetic geometry, and representation theory. The methodology relies on establishing a rigorous framework connecting the three invariants:

Tautological Subrings: The study is guided by the tautological subrings $R^*(M) \subset A^*(M)$ and $RH^*(M) \subset H^*(M)$ , defined as the smallest systems closed under pushforwards along gluing and forgetful maps. These are generated by "decorated boundary strata" involving $\psi$ and $\kappa$ classes.
Comparative Invariants: The paper utilizes the excision sequence for Chow groups, the cycle class map $cl: A^*(M) \to H^*(M)$ , and Deligne's weight spectral sequence to relate the cohomology of $M_{g,n}$ to that of $\overline{M}_{g,n}$ .
Arithmetic Tools: The Grothendieck–Lefschetz trace formula is employed to link point counts to the traces of the Frobenius endomorphism on compactly supported $\ell$ -adic cohomology. The paper distinguishes between "Tate type" cohomology (where eigenvalues are powers of $q$ ) and non-Tate cohomology (involving modular forms).
Equivariant Point Counts: To handle the boundary strata, which involve quotients by finite groups, the paper utilizes equivariant point counts and the Getzler–Kapranov formula, which relates the equivariant counts of $M_{g,n}$ to those of $\overline{M}_{g,n}$ .
Inductive Arguments: A key methodological tool is the use of inductive arguments on the genus and number of markings, often relying on the "Chow–Künneth generation Property" (CKgP) to decompose products of moduli spaces.

Key Contributions and Results

1. Relationships Between Invariants
The paper clarifies the conditions under which the invariants align.

If $A^*(\overline{M}_{g,n}) = R^*(\overline{M}_{g,n})$ and the space satisfies CKgP, then the cycle class map is an isomorphism, and $\#\overline{M}_{g,n}(\mathbb{F}_q)$ is a polynomial in $q$ .
Conversely, if $\#\overline{M}_{g,n}(\mathbb{F}_q)$ is a polynomial, the cohomology is pure Tate (vanishing odd cohomology and pure even cohomology), though this does not automatically imply the Chow ring is tautological without further assumptions.
For open moduli spaces $M_{g,n}$ , the relationship is more complex; non-Tate cohomology can exist even if the Chow ring is tautological, and point counts may fail to be polynomials due to odd-weight cohomology contributions.

2. Tautologicality of Chow and Cohomology Rings
The paper surveys the current state of knowledge regarding when $A^*(M)$ and $H^*(M)$ are generated by tautological classes, summarized via charts for $(g, n)$ :

Genus 0: $A^*(M_{0,n})$ is trivial; $A^*(\overline{M}_{0,n})$ is tautological.
Genus 1: $A^*(M_{1,n}) = R^*(M_{1,n})$ for $n \le 10$ . For $n \ge 11$ , $A^*(M_{1,n})$ is infinite-dimensional and non-tautological due to the existence of non-zero odd cohomology $H^{11}(M_{1,11})$ . However, the even cohomology of $M_{1,n}$ remains tautological.
Genus 2–9: Extensive results establish tautologicality for specific pairs $(g, n)$ (e.g., $g=2, n \le 9$ ; $g=3, n \le 9$ ; $g=7, n \le 3$ ). These results often rely on explicit stratifications (e.g., hyperelliptic vs. non-hyperelliptic loci) and models of curves as complete intersections.
Non-Tautological Classes: The paper identifies loci of double covers (bielliptic curves) as a primary source of non-tautological classes. For $g+m \ge 12$ , the class of the locus of bielliptic curves is non-tautological. This leads to the conclusion that $A^*(\overline{M}_{g,n}) \neq R^*(\overline{M}_{g,n})$ for $2g+n \ge 24$ (and higher bounds for specific $g$ ).
Cohomology in Low Weights: The paper details results on the vanishing of odd cohomology. $H^k(M_{g,n}) = 0$ for odd $k \le 9$ (except $H^{11}(M_{1,11})$ ). For $k=11, 13, 15$ , explicit descriptions of the cohomology as Galois representations are provided, often involving modular forms (e.g., the Ramanujan $\tau$ function for $k=11$ ).

3. Point Counts and Polynomiality

Polynomial Point Counts: The paper establishes that $\#\overline{M}_{g,n}(\mathbb{F}_q)$ is a polynomial in $q$ if and only if the cohomology is pure Tate. This is confirmed for $3g + 2n < 25$ .
Non-Polynomiality: For $M_{g,n}$ , point counts are generally not polynomials when odd cohomology is non-zero. The paper proves that $\#\overline{M}_{g,n}(\mathbb{F}_q)$ is not a polynomial for many cases where $2g+n \ge 24$ , driven by the non-vanishing of $\chi_k(M_{g,n})$ for odd $k$ .
Conjecture: The paper presents Conjecture 5.1: $\#\overline{M}_{g,n}(\mathbb{F}_q)$ is a polynomial in $q$ if and only if $3g + 2n < 25$ . This is proven for $n=0$ and computationally verified for $g+n < 150$ .
Approximate Polynomiality: Even when not strictly polynomial, point counts are well-approximated by polynomials, with corrections determined by coefficients of modular forms (e.g., Siegel modular forms).

Significance and Claims
The paper claims to provide a comprehensive, state-of-the-art survey of the interplay between the algebraic, topological, and arithmetic invariants of moduli spaces of curves. Its significance lies in:

Unification: It systematically connects the tautological nature of Chow rings, the structure of cohomology rings, and the arithmetic of point counts, showing how failures in one domain (e.g., non-tautological classes) manifest in others (e.g., non-polynomial point counts).
Clarification of Bounds: It delineates the precise ranges of $(g, n)$ where these spaces behave "nicely" (tautological rings, polynomial counts) versus where they exhibit complex behavior (infinite-dimensional Chow rings, non-Tate cohomology).
Synthesis of Recent Work: It consolidates results from various authors (including Canning, Payne, Faber, Bergström, and others) to resolve many open questions in low genera and provide a clear roadmap for the remaining unknown cases.
Evidence for Conjectures: It provides strong evidence for the Langlands program conjectures regarding the Galois representations appearing in the cohomology of these moduli spaces, specifically confirming Conjecture 2.20 for weights $k \le 15$ .

The paper does not claim to solve the general problem for all $g$ and $n$ , acknowledging that for large genera, the moduli spaces are of general type and current techniques (relying on unirationality or explicit stratifications) break down. Instead, it offers a detailed map of the known territory and the specific obstructions (such as bielliptic loci and odd cohomology) that prevent a complete description in the general case.