Dimension statistics of representations of finite groups

This paper investigates the dimension statistics of representations for finite groups, demonstrating that for reductive groups over finite fields and symmetric groups, quantities such as representation dimensions and conjugacy class sizes exhibit asymptotically constant or log-constant behavior as the field size or group rank tends to infinity.

The Gist

Imagine you are trying to understand the "personality" of a large group of people (a Finite Group). In mathematics, there are two main ways to describe this group:

1. The "Talent" View (Representations): How many different ways can this group perform? Each performance has a certain "size" or "complexity" (called dimension). If you square these sizes and add them all up, you get the total number of people in the group.
2. The "Gossip" View (Conjugacy Classes): How do people in this group interact? People who behave similarly (can be swapped with each other without changing the group's structure) form a "clique" or conjugacy class. If you count the size of each clique and add them up, you also get the total number of people.

The Big Question:
The authors, Arvind Ayyer and Dipendra Prasad, ask a "wishful" question: Are these two lists of numbers (the squared sizes of performances and the sizes of cliques) actually the same?

If you look at a small group like the symmetries of a triangle ($S_3$), the answer is no.

• Talent sizes: 1, 1, 2 (squared: 1, 1, 4). Sum = 6.
• Clique sizes: 1, 2, 3. Sum = 6.
  The numbers don't match up at all. It's like having a list of weights for apples and a list of weights for oranges; even if the total weight is the same, the individual items are different.

However, the paper explores what happens when the groups get huge (infinite families of groups). Do the lists start to look similar in a statistical sense?

The Main Discoveries

The paper divides its investigation into two main scenarios, using some clever analogies:

1. The "Smooth" Groups (Reductive Groups over Finite Fields)

Think of these groups as a massive, well-organized army where everyone follows strict rules. As the army gets bigger (either by adding more soldiers or making the rules more complex):

• The Finding: The "Talent" sizes and "Clique" sizes become statistically identical.
• The Analogy: Imagine a crowd of 1 million people. If you look at the height of every person, you might see a huge spread (some are 4 feet, some 7 feet). But if you look at a specific type of giant army, you might find that almost everyone is exactly 6 feet tall.
• The Result: For these groups, the "Talent" list and the "Clique" list are both "flat." Almost every performance has roughly the same size, and almost every clique has roughly the same size. They are "asymptotically constant." It's as if the group has forgotten its individual quirks and settled into a uniform rhythm.

2. The "Chaotic" Groups (Symmetric Groups)

Now, imagine the Symmetric Group ($S_n$). This is the group of all possible ways to shuffle a deck of $n$ cards. As $n$ gets huge, this group becomes incredibly chaotic and complex.

• The Finding: Here, the "Talent" sizes and "Clique" sizes are NOT the same. They are wildly different.
• The Analogy: Imagine a massive music festival.
  • The Talent View: Most bands are tiny (1 person), but a few are massive superstars (huge dimensions). The sizes are all over the place.
  • The Clique View: Most groups of people are small cliques, but a few are massive mobs.
  • The Twist: Even though both lists are "spread out," they are spread out in different ways. If you tried to match the biggest band to the biggest mob, they wouldn't line up.
• The Result: The authors prove that for these groups, the "Talent" list and the "Clique" list are orthogonal (at a 90-degree angle). In plain English: knowing the size of a performance tells you almost nothing about the size of the corresponding clique. They are completely uncorrelated.

Key Concepts Simplified

• Asymptotically Constant: As the group gets infinitely large, the data becomes boringly uniform. Everyone is the same size. (Happens with the "Smooth" groups).
• Asymptotically Log Constant: The data isn't uniform, but if you look at the logarithms (a way of squashing huge numbers down), they look uniform. This is a "middle ground" found in some cases.
• The 90-Degree Angle: In math, if two vectors are at a 90-degree angle, they are independent. The paper shows that for Symmetric Groups, the "Talent" vector and the "Clique" vector are at 90 degrees. They are totally unrelated.

Why Does This Matter?

The authors started with a "wishful thinking" that the two ways of counting a group (Talent vs. Clique) would be identical term-by-term. They found that:

1. For some groups (like GLn), the wish comes true in a statistical sense: the groups are so regular that the two lists merge into one.
2. For other groups (like Symmetric Groups), the wish fails spectacularly. The two lists are completely different, revealing a deep structural difference between how these groups "perform" and how they "interact."

In a Nutshell:
The paper is a statistical detective story. It asks, "If we look at the DNA of a group (its representations) and its social structure (its conjugacy classes), do they match?"

• For ordered, rigid groups, the answer is "Yes, they look the same."
• For chaotic, shuffling groups, the answer is "No, they are completely different."

This helps mathematicians understand the hidden "shape" of symmetry in the universe, distinguishing between groups that are predictable and those that are beautifully, chaotically complex.

Technical Summary

Here is a detailed technical summary of the paper "Dimension Statistics of Representations of Finite Groups" by Arvind Ayyer and Dipendra Prasad.

1. Problem Statement and Motivation

The paper investigates the statistical relationship between two fundamental quantities associated with a finite group $G$:

1. Dimension Data: The set of dimensions $\{d_\pi\}$ of its irreducible complex representations $\pi$.
2. Conjugacy Class Data: The set of sizes $\{|C_g|\}$ of its conjugacy classes.

The "Wishful Thinking" Hypothesis:
The authors explore a hypothetical scenario where the sum of squared dimensions equals the sum of conjugacy class sizes (both equal to $|G|$) not just globally, but term-by-term. Specifically, they ask if there exists a bijection between irreducible representations and conjugacy classes such that $d_\pi^2 \approx |C_g|$.

• Initial Observation: For small non-abelian groups (e.g., $S_3, S_4, D_8$), this term-by-term equality fails completely.
• Refined Question: Does this approximate equality hold in an asymptotic or statistical sense for large families of groups (e.g., as the field size $q \to \infty$ or the rank $n \to \infty$)?

2. Methodology

The authors employ a combination of representation theory, combinatorics, and asymptotic analysis. Their primary methodological tools include:

• Asymptotic Collinearity: They define vectors $d_G = (d_1, \dots, d_n)$ and $c_G = (\sqrt{|C_1|}, \dots, \sqrt{|C_n|})$ in $\mathbb{R}^n$. They analyze the angle $\theta$ between these vectors and the constant vector $\mathbf{1} = (1, \dots, 1)$.
  • If the angle between $d_G$ (or $c_G$) and $\mathbf{1}$ tends to 0 as $|G| \to \infty$, the data is termed asymptotically constant.
  • They also introduce asymptotically log constant, checking if the angle between $\ln(d_G)$ and $\mathbf{1}$ tends to 0.
• Kirillov Theory: For unipotent groups, they utilize the correspondence between irreducible representations and co-adjoint orbits to relate $d_\pi^2$ to orbit sizes.
• Generating Functions & Asymptotics: For symmetric groups, they use the Hardy-Ramanujan formula for partition functions, Stirling's approximation, and generating series techniques (including the "hybrid method" of Darboux and singularity analysis) to estimate sums of dimensions and class sizes.
• Vershik-Kerov Theory: They apply limit shape results for random partitions to understand the distribution of representation dimensions.

3. Key Contributions and Results

A. Finite Reductive Groups ($G(\mathbb{F}_q)$)

The paper establishes a strong statistical equivalence between dimension data and class sizes for reductive groups over finite fields.

• Varying $q$ (Fixed Rank): For a connected reductive group $G$ over $\mathbb{F}_q$, as $q \to \infty$:
  • The dimension data $\{d_\pi\}$ is asymptotically constant.
  • The conjugacy class sizes are also asymptotically constant.
  • Result: The angle between the dimension vector and the constant vector tends to 0. This implies that "most" irreducible representations have dimensions roughly equal to $q^{\text{rank}}$, and most conjugacy classes have sizes roughly equal to $q^{\text{rank}}$.
• Varying $n$ (Fixed $q$): For the general linear group $GL_n(\mathbb{F}_q)$ as $n \to \infty$:
  • The dimension data is asymptotically log constant but not asymptotically constant.
  • The limit of the normalized inner product involves a specific constant $\gamma(q) = \sum q^{-i(i+1)/2}$.

B. Finite Nilpotent Groups

• The authors introduce the concept of selfdual nilpotent groups (where the lower and upper central series coincide and quotients are isomorphic).
• They show that for specific selfdual $p$-groups, the dimension data $d_\pi^2$ matches the conjugacy class size data exactly, providing a class of groups where the "wishful thinking" holds true without asymptotic limits.

C. Symmetric Groups ($S_n$)

This is the most significant and counter-intuitive part of the paper. The behavior of $S_n$ differs fundamentally from reductive groups.

• Dimension Statistics:
  • Using the Plancherel measure (where weight $\propto d_\lambda^2$), the authors cite Vershik-Kerov and Bufetov to show that $\ln(d_\lambda^2)$ concentrates around a specific value proportional to $\sqrt{n}$.
  • However, under the uniform measure (equal weight for all partitions), the dimension data is asymptotically log constant but not asymptotically constant.
  • Crucial Finding: The angle between the dimension vector $(d_\lambda)$ and the constant vector $\mathbf{1}$ tends to $\pi/2$ as $n \to \infty$. This indicates a massive "spread" or dispersion in the dimensions; they are not concentrated around a single value.
• Conjugacy Class Statistics:
  • Using Erdős-Turan results and new estimates for the second moment of class sizes (Proposition 10.1), the authors prove that conjugacy class sizes exhibit the same statistical behavior as dimensions.
  • The class size data is also asymptotically log constant but not asymptotically constant, with the angle to the constant vector tending to $\pi/2$.
• Comparison: While $d_\lambda^2$ and $|C_\lambda|$ are not term-by-term equal, they share the same statistical "spread" and distribution properties in the limit.

4. Significance and Implications

1. Resolution of the "Wishful Thinking": The paper clarifies that while exact term-by-term equality ($d_\pi^2 = |C_g|$) rarely holds, a statistical equivalence exists. For reductive groups, the data is "flat" (concentrated), while for symmetric groups, the data is "spread out" but follows the same distributional laws for both dimensions and class sizes.
2. New Definitions: The introduction of asymptotically constant and asymptotically log constant provides a rigorous framework for comparing representation-theoretic data across different group families.
3. Symmetric Group Insights: The result that the angle between dimension vectors and the constant vector approaches $\pi/2$ for $S_n$ is a profound statement about the structure of symmetric groups. It implies that the "typical" representation dimension is not representative of the whole set; the distribution is heavy-tailed.
4. Connection to Number Theory: The use of moments (sums of $d_\lambda^k$) and generating functions bridges representation theory with analytic number theory (Hardy-Ramanujan, Stirling's formula), offering new tools for analyzing group representations.

Conclusion

The paper demonstrates that for large finite groups, the dimension data and conjugacy class size data are statistically "roughly" the same (up to squaring), but the nature of this similarity depends heavily on the group type. For reductive groups over large fields, the data is concentrated (constant). For symmetric groups and reductive groups with large rank, the data is dispersed (log constant), yet the dispersion patterns for dimensions and class sizes are identical, suggesting a deep, underlying duality in the structure of finite groups.