🔢 mathematics

On an entropic analogue of additive energy

This paper establishes $2{\bf H}\{X, Y\} - {\bf H}\{X+Y\}$ as a natural entropic analogue of additive energy, develops its foundational theory, applies it to prove Tao's entropy variant of the Balog--Szemerédi--Gowers theorem, and formulates sum-product conjectures for entropic energies in finite fields.

Original authors: Marcel K. Goh

Published 2026-04-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marcel K. Goh

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 a detective trying to understand how two groups of people interact. In the world of mathematics, these "groups" are usually sets of numbers, and the "interaction" is adding them together.

This paper, written by Marcel K. Goh, is about translating a very specific type of detective work from the world of counting numbers (Additive Combinatorics) into the world of measuring information (Information Theory).

Here is the breakdown of the paper using simple analogies.

1. The Two Worlds: Counting vs. Guessing

Imagine two different ways to describe a party:

The Combinatorial Way (Counting): You count exactly how many people are at the party ( $|A|$ ) and how many unique pairs of people you can form ( $|A+B|$ ). If you have 10 people and they can only form 15 unique pairs, something is very structured about this group.
The Entropic Way (Guessing): Instead of counting, imagine you are a guest trying to guess who is at the party. Entropy is a measure of how surprised you are.
- If everyone is wearing the same uniform, you aren't surprised at all (Low Entropy).
- If everyone is wearing a different, random outfit, you are very surprised (High Entropy).

The paper asks: Can we translate the rules of "counting pairs" into the rules of "guessing outfits"?

2. The Main Character: "Entropic Additive Energy"

In the old world of counting, mathematicians use a number called Additive Energy to measure how "redundant" a group is.

The Analogy: Imagine a group of people where many different pairs of people add up to the same total.
- Example: If $1+4 = 5$ and $2+3 = 5$ , that's a "collision."
- If a group has High Energy, it means there are lots of collisions. The group is very structured and predictable (like a grid).
- If a group has Low Energy, the sums are all unique. The group is chaotic and random (like a Sidon set).

The author introduces a new character: Entropic Additive Energy.
Instead of counting collisions, this measures the surprise of the sum.

Formula: $2 \times \text{Entropy}(X, Y) - \text{Entropy}(X+Y)$ .
Translation: If you know the individual people ( $X$ and $Y$ ) but the sum ( $X+Y$ ) is surprisingly predictable (low entropy), then the "Entropic Energy" is high. It means the group is structured.

3. The Big Discovery: The "Balog–Szemerédi–Gowers" Translation

There is a famous theorem in the counting world (the Balog–Szemerédi–Gowers theorem) that says:

"If a group has High Energy (lots of collisions), then there must be a smaller, hidden subgroup inside it that is very orderly."

The author proves that this rule works in the "Guessing" world too!

The Entropic Version: If your random variables have high "Entropic Energy," you can "condition" them (filter them based on the sum) to find a new pair of variables that are much more predictable and orderly.
Why it matters: This allows mathematicians to use the powerful tools of information theory (which are often more flexible) to solve hard problems about number sets.

4. The "Sidon Set" Mystery

The paper also looks at the opposite extreme: Sidon Sets.

The Analogy: Imagine a group where no two pairs add up to the same number. Every sum is unique. This is the most chaotic, "random" group possible.
The author shows that in the entropy world, these "Sidon" groups have a specific signature: their "doubling constant" (how much the group grows when you add it to itself) is at its maximum.
They even define a "Sidon Random Variable"—a variable that behaves like a Sidon set, even if the numbers it picks from aren't a perfect Sidon set.

5. The Final Puzzle: Sum vs. Product

The paper ends with a big conjecture about the Sum-Product Problem.

The Question: Can a group of numbers be "structured" in both addition and multiplication at the same time?
- Analogy: Can a group of people be perfectly organized by height (addition) AND perfectly organized by shoe size (multiplication) simultaneously?
- The Answer (in the old world): No. If they are organized by height, they must be messy by shoe size, and vice versa.
The Entropic Conjecture: The author proposes that this rule holds for information theory too. If a random variable is very predictable when you multiply its values, it must be very unpredictable when you add them (and vice versa), unless it's a very specific, boring case (like a sub-field).

Summary

This paper is a dictionary between two languages:

The Language of Sets: "How many ways can I add these numbers?"
The Language of Information: "How surprised am I by the result of adding these random variables?"

The author shows that the most important rules of the first language have perfect translations in the second. This is exciting because it lets mathematicians use the "surprise" metric to solve "counting" problems that were previously very difficult. It's like realizing that the rules of chess also apply to checkers, just with different pieces.

1. Problem Statement and Context

The paper addresses the growing field of analogies between additive combinatorics (dealing with sets, sumsets, and cardinalities) and information theory (dealing with random variables, sums of variables, and entropy). While many results in additive combinatorics have been successfully translated into the entropy setting (e.g., the polynomial Freiman–Ruzsa conjecture), a specific gap remained regarding the additive energy of sets.

Combinatorial Additive Energy: For two sets $A$ and $B$ , the additive energy $E(A, B)$ counts the number of quadruples $(a, a', b, b')$ such that $a+b = a'+b'$ . It measures the "redundancy" or structure within the sumset $A+B$ .
The Gap: While the concept of entropy $H(X)$ is well-established as an analogue to $\log |A|$ , a natural, rigorous definition of entropic additive energy $A\{X, Y\}$ that mirrors the properties and utility of $E(A, B)$ had not been fully developed or utilized in the literature, despite appearing implicitly in prior work (e.g., by Gowers, Green, Manners, and Tao).

The paper aims to define this quantity, establish its fundamental properties, and demonstrate its utility in simplifying proofs of major theorems, specifically Tao's entropy variant of the Balog–Szemerédi–Gowers theorem.

2. Methodology and Definitions

The author employs an axiomatic approach based on the Khintchine–Shannon axioms for entropy (invariance, extensibility, continuity, maximization, additivity, and normalization).

Definition of Entropic Additive Energy

Let $X$ and $Y$ be discrete random variables taking values in the same abelian group.

Construction: Let $(X_1, Y_1)$ and $(X_2, Y_2)$ be conditionally independent trials of $(X, Y)$ relative to their sum $S = X+Y$ . This means $X_1+Y_1 = X_2+Y_2 = X+Y$ , and given the sum, the pairs are independent.
Definition: The entropic additive energy is defined as:
$A\{X, Y\} = H\{X_1, Y_1, X_2, Y_2\}$
Simplified Formula: Using the chain rule and conditional independence, this reduces to a more practical form:
$A\{X, Y\} = 2H\{X, Y\} - H\{X + Y\}$
(Note: If $X$ and $Y$ are independent, $A\{X, Y\} = 2H\{X\} + 2H\{Y\} - H\{X+Y\}$ .)

Entropic Multiplicative Energy

Similarly, for variables in a ring (specifically a field), the entropic multiplicative energy is defined as:
$M\{X, Y\} = 2H\{X, Y\} - H\{XY\}$

3. Key Contributions and Results

A. Fundamental Properties and Analogies

The paper proves that $A\{X, Y\}$ behaves analogously to the combinatorial $E(A, B)$ :

Bounds: For uniform variables $U_A, U_B$ on sets $A, B$ , the paper establishes:
$\log \frac{|A|^2|B|^2}{|A+B|} \leq A\{U_A, U_B\} \leq \log E(A, B)$
Inequalities: The author demonstrates that classical inequalities like Cauchy-Schwarz do not always translate directly. For instance, the inequality $E(A, B) \leq E(A, A)^{1/2}E(B, B)^{1/2}$ fails in the entropy setting, providing a counterexample using specific integer sets.
Large Energy Regime: The paper shows that large $A\{X, Y\}$ implies that the sum $X+Y$ has "small" entropy relative to the individual entropies, provided $X$ and $Y$ are not too dependent.

B. Entropic Balog–Szemerédi–Gowers Theorem

A major contribution is the reformulation and simplification of Tao's Entropic Balog–Szemerédi–Gowers Theorem.

Classical Theorem: If sets $A, B$ have high additive energy, there exist large subsets $A' \subseteq A, B' \subseteq B$ such that $|A'+B'|$ is small.
Entropic Version: If $A\{X, Y\}$ $A {X, Y}$ is large (specifically $A\{X, Y\} \geq \frac{3}{2}H\{X\} + \frac{3}{2}H\{Y\} - \log C$ $A {X, Y} \geq \frac{3}{2} H {X} + \frac{3}{2} H {Y} - lo g C$ ), then there exist conditionings $X', Y'$ $X^{'}, Y^{'}$ of $X, Y$ $X, Y$ relative to a random variable $Z$ $Z$ (where $Z=X+Y$ $Z = X + Y$ ) such that:
1. $X'$ and $Y'$ are conditionally independent given $Z$ .
2. Their conditional entropies $H\{X'|Z\}$ and $H\{Y'|Z\}$ are close to the original entropies.
3. The conditional entropy of the sum $H\{X'+Y'|Z\}$ is small.
Significance: The author proves this using the new definition of $A\{X, Y\}$ , which streamlines the proof compared to Tao's original formulation, making the link to the classical theorem more transparent.

C. Small Energy Regime and Sidon Sets

The paper analyzes the regime where entropic additive energy is small.

Small Energy Implication: Small $A\{X, Y\}$ corresponds to $H\{X+Y\}$ being large (close to $H\{X\} + H\{Y\}$ ), indicating "random-like" behavior.
Sidon Sets: The author connects small energy to Sidon sets (sets where all pairwise sums are distinct).
- A random variable $X$ is defined as "Sidon" if its entropic doubling constant $s\{X\} = H\{X+X'\} - H\{X\}$ satisfies $s\{X\} \geq H\{X\} - 1$ .
- The paper proves that if $X$ is supported on a Sidon set, it satisfies this bound. Conversely, it provides examples of non-Sidon sets that still yield "Sidon-like" random variables, showing the entropy definition is strictly more general than the combinatorial set definition.

D. Sum-Product Conjectures

The final section extends the framework to multiplicative energy and formulates entropic analogues of the Sum-Product Problem.

Context: In classical additive combinatorics, the Bourgain–Katz–Tao theorem states that a set cannot have both small sumset and small product set (unless it is a subfield).
Conjecture 14 (Entropic BKT): For a random variable $X$ in a finite field $\mathbb{F}_p$ with entropy in a specific range, if the multiplicative energy $M\{X\}$ is large, then the additive energy $A\{X\}$ must be small (and vice versa).
Proof Strategy: The author shows that this conjecture follows from an Entropic Glibichuk–Konyagin Lemma (a sum-product type lemma for entropy).
Real Line Conjecture: The paper proposes an entropic version of the Erdős–Szemerédi conjecture, suggesting that for any random variable on $\mathbb{R}$ , $\max(A\{X\}, M\{X\}) \leq (3-\epsilon)H\{X\}$ . The author notes that recent independent work (Li et al.) suggests the optimal $\epsilon$ in this setting is bounded by $1/3$ , limiting the ability to improve classical exponents via this entropy method.

4. Significance and Impact

Unification: The paper successfully unifies the concepts of additive energy and entropy, providing a rigorous definition ( $A\{X, Y\} = 2H\{X, Y\} - H\{X+Y\}$ ) that preserves the structural intuition of the combinatorial case.
Simplification: It simplifies the proof of the Entropic Balog–Szemerédi–Gowers theorem, a cornerstone result in the field, by leveraging the properties of the newly defined energy.
New Directions: By defining entropic Sidon sets and formulating entropic sum-product conjectures, the paper opens new avenues for research in information-theoretic additive combinatorics.
Clarification of Limits: The paper carefully distinguishes where the entropy analogy holds and where it diverges from the combinatorial world (e.g., the failure of certain Cauchy-Schwarz analogues and the specific bounds on the sum-product exponent), providing a realistic map of the field's capabilities.

In summary, Goh's work provides the necessary theoretical infrastructure to treat "entropic additive energy" as a first-class object, enabling cleaner proofs of existing theorems and the formulation of new conjectures in the intersection of combinatorics and information theory.