Fourier dimension of imaginary Gaussian multiplicative chaos

This paper establishes that the Fourier dimension of imaginary Gaussian multiplicative chaos on the unit circle in the subcritical phase is almost surely $1-\beta^2$, while also proving its failure to belong to a critical Sobolev space and demonstrating that its high-frequency coefficients converge to independent complex Gaussians, effectively behaving as white noise.

The Gist

Imagine you are standing in a vast, foggy room. The fog isn't uniform; it's made of tiny, swirling particles that dance around in a chaotic, unpredictable way. In mathematics, this "fog" is called Gaussian Multiplicative Chaos. It's a way of describing a random, jumbled field of energy that exists everywhere but is impossible to pin down at any single point.

Usually, when mathematicians study this fog, they look at it as a "positive" thing—like a pile of sand or a cloud of gas. But in this paper, the authors look at a very specific, strange version of this fog: the Imaginary version.

Think of the "Real" fog as a pile of sand you can weigh. The "Imaginary" fog is more like a ghostly, vibrating melody. It doesn't have weight; it has a phase and a frequency. It's a complex, swirling sound wave that exists in the air but can't be touched.

The Big Question: How "Rough" is the Sound?

The authors wanted to answer a specific question about this ghostly melody: How fast does the sound fade away as you listen to higher and higher pitches?

In music, low notes are deep and rumbling. High notes are sharp and thin. If you take a recording of this chaotic "imaginary fog" and break it down into its individual notes (its Fourier coefficients), the authors wanted to know: How quickly do the high notes disappear?

They found a precise rule. If you control the "intensity" of the chaos with a number called $\beta$ (beta), the high notes fade away at a speed determined by the formula $1 - \beta^2$.

• The Analogy: Imagine the fog is a piece of fabric. If the fabric is very rough (high $\beta$), the high-frequency ripples (the tiny wrinkles) die out very fast. If the fabric is smoother (low $\beta$), the ripples last longer. The authors proved that the "roughness" of this imaginary fabric is exactly predictable.

The "White Noise" Surprise

Here is the most magical part of their discovery.

Usually, when you have a chaotic system, the different parts of the noise are tangled together. If you hear a loud note, it might influence the next note. But the authors found that if you look at this imaginary fog at very high frequencies, it behaves like White Noise.

• The Analogy: Imagine listening to a radio tuned between stations. You hear a hiss. That hiss is "white noise"—it's random, and every tiny sound is completely independent of the one before it.
• The paper proves that if you take this complex, swirling imaginary chaos and zoom in on the highest frequencies, it stops looking like a structured, complex wave and starts looking exactly like that random radio hiss. The "notes" become independent, random strangers, each with no memory of the others.

How Did They Solve It? (The Secret Weapon)

You might wonder, "How do you calculate the behavior of a ghostly, infinite fog?"

The authors used a very old, very powerful mathematical tool called Jack Polynomials.

• The Analogy: Think of Jack Polynomials as a special set of Lego bricks. Usually, building with these bricks is incredibly hard because they snap together in complex, unpredictable ways.
• However, the authors discovered that when you build with these bricks at a very specific scale (the "large gap" regime), the bricks suddenly become simple. They stop snapping together in complex patterns and just stack up in a straight line.
• By realizing that the complex math simplifies into a straight line when you look at the highest frequencies, they were able to count the pieces and prove exactly how the noise behaves.

What About "Real" Fog?

The paper also mentions that this result is robust. Even if you slightly change the rules of the fog (adding a little bit of smoothness or changing the background texture), the main rule ($1 - \beta^2$) still holds true. It's like saying that no matter how you slightly tweak the recipe for a cake, the way it rises in the oven remains the same.

Summary of the Findings

1. The Dimension: They proved that the "Fourier dimension" (a measure of how fast the high notes fade) of this imaginary chaos is exactly $1 - \beta^2$.
2. The Limit: As you go to higher and higher frequencies, the chaos stops being a complex, tangled wave and turns into pure, independent random noise (White Noise).
3. The Method: They used a deep connection between random chaos and a specific type of mathematical symmetry (Jack Polynomials) to turn a messy problem into a clean, solvable one.

In short, the paper tells us that even in the most chaotic, imaginary, and ghostly mathematical worlds, there is a hidden, simple order waiting to be found if you look at the right frequency.

Technical Summary

Technical Summary: Fourier Dimension of Imaginary Gaussian Multiplicative Chaos

Problem Statement
This paper investigates the high-frequency Fourier asymptotics of imaginary Gaussian multiplicative chaos (GMC) on the unit circle $\mathbb{T}$. The object of study is the complex-valued random distribution formally defined as $M_{i\beta} = \exp(i\beta X)$, where $X$ is a log-correlated Gaussian field with covariance $E[X_\theta X_{\theta'}] = \log \frac{1}{|e^{i\theta} - e^{i\theta'}|}$ and $\beta \in (0, 1)$ is the subcritical parameter.

Unlike real GMC, which yields a positive multifractal measure, imaginary GMC is a proper random distribution that does not converge to a measure. Consequently, standard geometric notions of dimension do not apply directly. The authors focus on the Fourier dimension, defined as the optimal polynomial decay exponent $s$ such that $|\hat{\phi}(n)|^2 = O(|n|^{-s})$. While previous regularity results established that $M_{i\beta}$ belongs to Sobolev spaces $H^s(\mathbb{T})$ for $s < -\beta^2/2$ (implying an upper bound of $1-\beta^2$ for the Fourier dimension), the exact value of the Fourier dimension and the precise statistical behavior of the Fourier coefficients at high frequencies remained open.

Methodology
The authors employ the method of moments combined with the integrable structure provided by the exact logarithmic kernel. The core technical machinery involves:

1. Coulomb-gas Integrals: The moments of the Fourier coefficients, $E[|\hat{M}_{i\beta}(n)|^{2N}]$, are represented as integrals over the torus, resembling partition functions of two-component Coulomb gases.
2. Jack Polynomial Expansions: Using Stanley's Cauchy identity, the authors expand the interaction terms in the integrals using Jack polynomials (orthogonal polynomials associated with the Selberg inner product). This reduces the moment calculations to explicit sums over integer partitions.
3. Asymptotic Analysis of Partitions: The asymptotic behavior of these sums as the frequency $n \to \infty$ is analyzed by focusing on the "large gap regime," where the gaps between consecutive parts of the partition tend to infinity. In this regime, the Pieri coefficients (which arise from multiplying Jack polynomials by elementary symmetric polynomials) simplify asymptotically to 1.
4. Robustness via Coupling: To extend results beyond the exactly log-correlated field, the authors utilize a spectral coupling argument. They decompose a perturbed field $X_g$ into the standard log-correlated field $X$ plus a smoother Gaussian remainder $Y$. This allows them to transfer decay properties from the standard case to the perturbed case via convolution arguments in Sobolev spaces.

Key Contributions and Results

1. Exact Fourier Dimension:
   The primary result (Theorem 1.1) proves that for $\beta \in (0, 1)$, the Fourier dimension of $M_{i\beta}$ is almost surely equal to $1 - \beta^2$. This establishes the sharp lower bound, confirming that the decay is exactly as predicted by the upper bound derived from Sobolev regularity.

2. Critical Sobolev Regularity:
   The authors prove (Corollary 1.2) that $M_{i\beta}$ almost surely does not belong to the critical Sobolev space $H^{-\beta^2/2}(\mathbb{T})$. This resolves the open question regarding the boundary of regularity for imaginary chaos.

3. Robustness under Perturbations:
   The result regarding the Fourier dimension is shown to be robust (Theorem 1.3). It holds for a broad class of log-correlated fields where the covariance differs from the exact logarithmic kernel by a sufficiently regular function $g$ (specifically $g \in H^{1+\delta}$ in the stationary case or $g \in H^{3/2+\delta}$ generally).

4. Central Limit Theorem for Coefficients:
   For the exactly log-correlated case, the paper establishes a Central Limit Theorem (Theorem 1.4). The rescaled Fourier coefficients $n^{(1-\beta^2)/2} \hat{M}_{i\beta}(n)$ converge in law to an isotropic complex Gaussian random variable with variance $\kappa(\beta) = \frac{1}{\pi} \Gamma(1-\beta^2) \sin(\frac{\pi\beta^2}{2})$. Furthermore, finitely many consecutive coefficients converge jointly to independent copies of this variable.

5. Convergence to Complex White Noise:
   The high-frequency content of the chaos behaves asymptotically as white noise. Specifically, the rescaled and shifted distribution $n^{(1-\beta^2)/2} e^{in\theta} M_{i\beta}$ converges in distribution in $H^s(\mathbb{T})$ (for $s < -1/2$) to a complex white noise with intensity $\kappa(\beta)$ (Theorem 1.5).

Significance and Claims
The paper claims to provide the first precise determination of the Fourier dimension for imaginary Gaussian multiplicative chaos, a regime where standard techniques relying on positivity and multifractal mass estimates (used for real GMC) fail. By leveraging the specific integrable structure of the imaginary case via Jack polynomials, the authors demonstrate that the high-frequency content of the chaos is statistically indistinguishable from white noise, despite the underlying field being highly correlated.

The authors note that while the exact moment identities rely on the specific logarithmic kernel, the Fourier dimension result extends to perturbed fields. However, the Central Limit Theorem and the precise white noise convergence are currently established only for the exactly log-correlated case, as the factorization required for the general case involves complex asymptotic dependencies that are not fully resolved in this work. The paper also posits a conjecture that for general complex parameters $\gamma = \alpha + i\beta$, the Fourier dimension should be $\dim_F(M_\alpha) - \beta^2$.