Abstract
We consider random analytic functions given by a Taylor series with independent, centered complex Gaussian coefficients. We give a new sufficient condition for such a function to have bounded mean oscillations. Under a mild regularity assumption this condition is optimal. Using a theorem of Holland and Walsh, we give as a corollary a new bound for the norm of a random Gaussian Hankel matrix. Finally, we construct some exceptional Gaussian analytic functions which in particular disprove the conjecture that a random analytic function with bounded mean oscillations always has vanishing mean oscillations.
Introduction
In complex analysis, the regularity of a function is often assessed through the behavior of its Taylor coefficients. A central object of study is the Gaussian Analytic Function (GAF), where coefficients are independent complex Gaussian random variables. A well-known phenomenon in this field is the "regularity boost": a power series with random coefficients is almost always better behaved than a deterministic one with identical coefficient magnitudes.
This research focuses on characterizing the membership of GAFs in two specific function spaces:
- BMOA (Bounded Mean Oscillation): Functions that do not deviate excessively from their average value over any interval on the boundary.
- VMOA (Vanishing Mean Oscillation): A subset of BMOA where these oscillations effectively disappear as one zooms in on smaller scales.
The paper addresses the exact threshold for coefficients that places a GAF into BMOA and investigates whether a random function in BMOA necessarily belongs to the VMOA category.
Methodology
The authors employ a synthesis of high-dimensional probability and classical harmonic analysis:
- Dyadic Decomposition: Taylor coefficients are grouped into "blocks" (dyadic blocks) to analyze the total variance ($\sigma_k^2$) of these blocks, simplifying the infinite series into a manageable sequence of variances.
- Sledd Spaces (SL): The authors utilize a specific function space introduced by Sledd in 1981. This space serves as a powerful proxy for proofs as it is contained within BMOA but is more analytically tractable.
- Probabilistic Concentration: To manage randomness, the study employs Talagrand’s technique for controlling the supremum of stochastic processes and Gaussian concentration inequalities. These tools allow the authors to prove that if a property holds "on average," it almost certainly holds for the random function.
- Fejér Kernels: These classical tools from Fourier analysis are used to approximate functions and prove the approximability of the norms under study.
Key Insights and Results
Characterization of BMOA and VMOA
The authors demonstrate that if the variances of the coefficient blocks are decreasing (dyadic-regular), the condition for BMOA membership is remarkably simple. They prove that a dyadic-regular GAF belongs to BMOA (and even VMOA) if and only if the sequence of its Taylor coefficients is square-summable ($\ell^2$).
Disproving the VMOA Conjecture
It was long suspected that the inherent randomness of GAFs might force any BMOA function to automatically reside in VMOA. The authors disprove this by constructing "exceptional" GAFs. By strategically "piling up" randomness at specific scales, they maintain a constant level of oscillation, preventing the function from "vanishing" at the limit. This establishes a strict hierarchy of inclusions: $$SL_G \subsetneq VMOA_G \subsetneq BMOA_G \subsetneq (H^2 \cap \mathcal{B})_G$$ This proof confirms that these function spaces remain truly distinct even in a stochastic setting.
Bounds for Random Hankel Matrices
The research bridges function theory and linear algebra by viewing a GAF as the "symbol" of a Hankel matrix (a matrix where each ascending skew-diagonal is constant). Using their BMOA results, the authors provide a new upper bound for the operator norm of an $n \times n$ random Gaussian Hankel matrix $A$: $$E\|A\|^2 \le C \sum_{k=0}^L \sup_{k \le m \le L} \sigma_m^2$$ where $L \approx \log_2(2n)$. This bound is sharper than previous estimates for many types of coefficient sequences.
Conclusion and Future Work
While this work provides a sharp characterization of BMOA for GAFs, it leaves open the search for a "natural separable subspace" $S$ of BMOA that perfectly captures the behavior of Gaussian Analytic Functions (i.e., where $BMOA_G = S_G$). Additionally, while the focus here is on Gaussian distributions, the authors suggest that these results could likely be generalized to other probability distributions, providing a fertile ground for future research.
Summary updated