Random Chowla's Conjecture for Rademacher Multiplicative Functions

This paper proves that the partial sums of Rademacher random multiplicative functions evaluated at specific polynomial arguments converge in distribution to a standard Gaussian, confirming a conjecture by Najnudel and establishing that the fluctuations of sums over quadratic arguments almost surely reach the magnitude predicted by the law of the iterated logarithm.

The Gist

Imagine you are trying to predict the weather. You know that individual raindrops fall randomly, but if you look at a massive storm system, patterns emerge. In mathematics, specifically in the world of number theory, there is a famous mystery about a function called the Möbius function. This function acts like a cosmic coin flip for numbers: it assigns a value of +1 or -1 to certain numbers based on their prime factors.

For over a century, mathematicians have wondered: Does this function behave like a truly random coin flip, or is there a hidden pattern?

This paper, written by Jake Chinis and Besfort Shala, takes a bold step to answer that question by creating a "simulated universe" of numbers. Here is the breakdown of their discovery in simple terms.

1. The Setup: The "Magic Dice"

The authors imagine a world where the Möbius function is replaced by a Rademacher Random Multiplicative Function.

• The Analogy: Imagine you have a giant bag of dice. For every prime number (2, 3, 5, 7...), you roll a die. If it's even, you write down +1; if odd, -1.
• The Rule: Once you decide the value for a prime, you must follow a strict rule for all its multiples. If the prime 2 is +1, then any number made of 2s (like 4, 8, 16) gets a value based on multiplying those +1s together.
• The Result: You get a sequence of numbers that looks chaotic and random, but follows the hidden "multiplicative" rules of the universe.

2. The Main Discovery: The "Random Walk"

The authors asked: If we add up these random numbers for a specific type of input (like plugging numbers into a polynomial equation, e.g., $n^2 + 1$), what happens?

• The Old Belief: For a long time, people thought these sums might behave strangely or cancel out in weird ways.
• The New Proof: The authors proved that for most polynomial inputs (like $n^2+1$ or $(n+1)(n+2)$), the sum behaves exactly like a drunkard's walk.
  • The Metaphor: Imagine a drunk person walking down a street. They take a step forward (+1) or backward (-1) at random. If you watch them for a long time, they will wander away from the starting point, but their position will follow a perfect Bell Curve (the classic "Normal Distribution").
  • The Result: The paper confirms that these random number sums do form a Bell Curve. This proves that, in this random model, the numbers are behaving with "perfect" pseudo-randomness. This settles a long-standing guess made by mathematician Najnudel.

3. The "Big Fluctuations": The Tsunami

After proving the average behavior is a Bell Curve, the authors looked at the extremes.

• The Question: How far can the drunkard wander from the center? Is there a limit?
• The Law of the Iterated Logarithm: In pure probability, there is a rule that says a random walker will occasionally take a massive step, reaching a distance roughly equal to $\sqrt{N \log \log N}$ (where $N$ is the number of steps).
• The Finding: The authors showed that for the specific polynomial $n^2 + 1$, these random sums do occasionally hit these massive "tsunami" levels. They proved that almost surely, there will be moments where the sum is huge, matching the theoretical maximum expected for a random walk.

4. Why Does This Matter?

You might ask, "Why do we care about a fake random function?"

• The Bridge: The real Möbius function (the one in the real world) is deterministic, not random. We don't know if it behaves randomly or not.
• The Insight: By proving that the random model behaves exactly as we expect a truly random system to behave (forming Bell Curves and hitting specific fluctuation limits), the authors provide a strong "sanity check."
• The Takeaway: If the real Möbius function behaves differently than this random model, it would mean there is a deep, hidden structure in the primes that we haven't found yet. If it behaves the same, it suggests the primes are as chaotic as a coin flip. This paper confirms that the random model is robust and reliable, giving mathematicians a better tool to study the real world.

Summary

In short, Chinis and Shala built a mathematical simulation of the universe's prime numbers. They showed that when you plug these simulated numbers into polynomial equations, they dance to the rhythm of a perfect random walk. They proved that these sums follow the famous Bell Curve and occasionally jump to the massive heights predicted by probability theory. This gives us confidence that our understanding of "randomness" in number theory is on the right track.

Technical Summary

1. Problem Statement

The paper investigates the statistical distribution of partial sums of Rademacher random multiplicative functions (RMFs) evaluated at polynomial arguments.

• Context: The study of the Möbius function $\mu(n)$ is central to analytic number theory. The Riemann Hypothesis is equivalent to the bound $\sum_{n \le x} \mu(n) \ll x^{1/2+\epsilon}$. A heuristic model treats $\mu(n)$ as a sequence of independent random variables taking values $\pm 1$ (Rademacher) or on the unit circle (Steinhaus).
• The Conjecture: The authors address the Random Chowla Conjecture. While the deterministic Chowla conjecture concerns the autocorrelation of $\mu(n)$, the random version asks whether the partial sums of a random multiplicative function $f(P(n))$ behave like a random walk.
• Specific Gap: Previous work by Najnudel, Klurman, Shkredov, and Xu established that for Steinhaus RMFs (complex-valued, completely multiplicative), the normalized sums converge to a standard complex Gaussian. However, the case for Rademacher RMFs (real-valued, supported on square-free integers) was open. The Rademacher case is technically more difficult because the function is not completely multiplicative and is restricted to square-free values, introducing arithmetic constraints (square-free conditions) into the moment calculations.
• Goal: Prove that for specific polynomials $P(x)$, the normalized sum $\frac{1}{\sqrt{N}} \sum_{n \le N} f(P(n))$ converges in distribution to a standard real Gaussian $\mathcal{N}(0,1)$. Additionally, the paper investigates large fluctuations (Law of the Iterated Logarithm) for the specific case $P(n) = n^2 + 1$.

2. Methodology

The authors employ a combination of probabilistic limit theorems and deep arithmetic geometry techniques.

A. Probabilistic Framework: McLeish's Central Limit Theorem

The core probabilistic tool is McLeish's CLT for martingale difference sequences (Lemma 2.1).

• The authors decompose the sum $M_N = \sum_{n \le N} f(P(n))$ based on the largest prime factor of the arguments.
• They define a filtration based on primes and show that the partial sums form a martingale difference sequence.
• To apply the CLT, they must verify three conditions:
  1. Normalized Variances: The sum of variances converges to 1.
  2. Lindeberg Condition: (Often replaced by a stronger fourth-moment condition).
  3. Cross-terms Condition: The correlation between different components vanishes.
• Crucial Reduction: Verifying these conditions reduces to counting the number of solutions to specific Diophantine equations involving the polynomial $P(n)$. Specifically, they need to estimate the second moment (counting square-free values) and the fourth moment (counting quadruples where the product of four polynomial values is a perfect square).

B. Arithmetic Geometry: Counting Integral Points

The technical heart of the paper lies in estimating the number of solutions to:
$$P(n_1)P(n_2)P(n_3)P(n_4) = \square$$
where each $P(n_i)$ is square-free.

• Square-free Density: They utilize results (Lemma 2.6) on the density of square-free values of polynomials, relying on work by Ricci, Mirsky, and Tsang.
• Curve Counting: The problem of counting non-trivial solutions to $P(n_1)P(n_2) = P(n_3)P(n_4)$ (or the fourth moment equivalent) is transformed into counting integral points on curves of the form $aP(x) - bP(y) = 0$.
• Key Tools:
  • Bombieri-Pila Theorem: Used to bound integral points on absolutely irreducible curves.
  • Cilleruelo-Garaev / Granville: Used for quadratic curves and handling specific factorization cases.
  • Bootstrapping Argument: A novel technique developed to handle the case where the greatest common divisor (GCD) of polynomial values is large. This allows them to reduce the counting problem to curves of lower degree or smaller height.

C. Large Fluctuations (Law of Iterated Logarithm)

For the second main result (Theorem 1.3), the authors adapt the strategy used by Harper and Klurman-Shkredov-Xu for Steinhaus RMFs:

• Localization: They identify scales $x_i$ where $n^2+1$ has a very large prime factor $p > x_i \log x_i$.
• Conditioning: They condition on the values of $f(q)$ for small primes, isolating the contribution of the large prime factors.
• Normal Approximation: The remaining sum is viewed as a weighted sum of independent random variables. They use a multivariate normal approximation lemma (Lemma 5.1) to show that the maximum of these sums over multiple scales behaves like the maximum of independent Gaussians, yielding the $\sqrt{\log \log N}$ fluctuation.

3. Key Contributions and Results

Theorem 1.2: Rademacher Random Chowla Conjecture

• Statement: Let $f$ be a Rademacher RMF and $P \in \mathbb{Z}[x]$ be either:
  1. A product of at least two distinct linear factors, OR
  2. An irreducible quadratic polynomial satisfying an admissibility condition (no prime $p$ such that $p^2 | P(n)$ for all $n$).
     Then, there exists a constant $\kappa_P > 0$ such that:
     $$\frac{1}{\sqrt{\kappa_P N}} \sum_{n \le N} f(P(n)) \xrightarrow{d} \mathcal{N}(0, 1)$$
• Significance: This confirms a conjecture by Najnudel and resolves a question posed by Klurman, Shkredov, and Xu. It bridges the gap between the complex (Steinhaus) and real (Rademacher) cases, showing that the pseudo-random behavior holds even with the stricter square-free constraint.

Proposition 3.1: Fourth Moment Asymptotic

• Statement: For the polynomials described above, the number of quadruples $(n_1, n_2, n_3, n_4) \le N$ such that $P(n_1)P(n_2)P(n_3)P(n_4)$ is a perfect square (with each factor square-free) is asymptotically $3\kappa_P^2 N^2 + O(N^{2-\delta})$.
• Significance: This proves that the "diagonal" solutions (where indices pair up) dominate the fourth moment, while "off-diagonal" solutions are negligible. This is the critical step required to satisfy the fourth-moment condition of McLeish's CLT. The proof introduces a bootstrapping argument to handle large GCDs of polynomial values, a technique not present in the Steinhaus case.

Theorem 1.3: Large Fluctuations

• Statement: For $P(n) = n^2 + 1$, there almost surely exist arbitrarily large $N$ such that:
  $$\left| \sum_{n \le N} f(n^2 + 1) \right| \gg \sqrt{N \log \log N}$$
• Significance: This matches the lower bound predicted by the Law of the Iterated Logarithm (LIL). It suggests that the random model accurately captures the extreme fluctuations expected in the deterministic Möbius function (under the assumption that the random model is a good proxy for $\mu$ on polynomials).

4. Significance and Impact

1. Resolution of Open Problems: The paper settles the Rademacher case of the Random Chowla Conjecture for a broad class of polynomials, a problem that had remained open despite significant progress in the Steinhaus case.
2. Technical Innovation: The development of the bootstrapping argument for counting integral points on curves arising from polynomial GCDs is a significant methodological advance. It allows for the treatment of the "square-free" constraint which complicates the Rademacher case compared to the Steinhaus case.
3. Connection to Deterministic Number Theory: By establishing Gaussian behavior and LIL-type fluctuations for the random model, the authors provide strong heuristic evidence for the behavior of the deterministic Möbius function on polynomial sequences. Specifically, they support the belief that $\sum \mu(n^2+1)$ behaves like a random walk with $\sqrt{N \log \log N}$ fluctuations, contrasting with the $\sqrt{N}$ fluctuations of the standard Möbius sum (if RH is true).
4. Future Directions: The paper notes that the upper bound for the LIL and the case of products of linear factors for general polynomials remain open, setting the stage for future research in probabilistic number theory.

In summary, this paper successfully extends the probabilistic understanding of multiplicative functions from the complex (Steinhaus) to the real (Rademacher) setting, utilizing sophisticated tools from analytic number theory and arithmetic geometry to overcome the specific difficulties posed by square-free constraints.