[Paper Review] Spectral Norm of Symmetric Functions

Anil Ada, Omar Fawzi+1 more|arXiv (Cornell University)|May 23, 2012

Coding theory and cryptography15 references4 citations

TL;DR

This paper provides a combinatorial characterization of the spectral norm of symmetric Boolean functions, showing that its logarithm is tightly bounded by $ r(f)\log(n/r(f)) $, where $ r(f) $ captures the minimal interval around the middle levels where the function or its parity-shifted version becomes constant. The result yields tight bounds on decision tree and communication complexity for symmetric functions and supports key conjectures in communication complexity and learning theory.

ABSTRACT

The spectral norm of a Boolean function $f:\{0,1\}^n o \{-1,1\}$ is the sum of the absolute values of its Fourier coefficients. This quantity provides useful upper and lower bounds on the complexity of a function in areas such as learning theory, circuit complexity, and communication complexity. In this paper, we give a combinatorial characterization for the spectral norm of symmetric functions. We show that the logarithm of the spectral norm is of the same order of magnitude as $r(f)\log(n/r(f))$ where $r(f) = \max\{r_0,r_1\}$, and $r_0$ and $r_1$ are the smallest integers less than $n/2$ such that $f(x)$ or $f(x) \cdot parity(x)$ is constant for all $x$ with $\sum x_i \in [r_0, n-r_1]$. We mention some applications to the decision tree and communication complexity of symmetric functions.

Motivation & Objective

• To provide a combinatorial characterization of the spectral norm for symmetric Boolean functions.
• To establish tight bounds on the spectral norm in terms of structural parameters of the function.
• To apply the characterization to decision tree complexity and communication complexity of symmetric functions.
• To support and extend conjectures in communication complexity and learning theory involving spectral norms.
• To provide a foundation for understanding the approximate monomial complexity and communication complexity of xor functions.

Proposed method

• Define $ r_0 $ and $ r_1 $ as the smallest integers less than $ n/2 $ such that $ f $ or $ f \cdot \text{parity} $ is constant on inputs with Hamming weight in $[r_0, n - r_1]$.
• Introduce $ r(f) = \max\{r_0, r_1\} $ as a key structural parameter of the function.
• Establish that $ \log \|\widehat{f}\|_1 = \Theta^*\left(r(f) \log(n / r(f))\right) $, where $ \|\widehat{f}\|_1 $ is the spectral norm.
• Use Fourier analytic techniques and duality to relate spectral norm to the complexity of parity decision trees and communication protocols.
• Leverage known results from Bruck and Smolensky on low-spectral-norm functions being representable as low-degree polynomials.
• Apply the characterization to analyze the approximate rank and communication complexity of xor functions $ f(x \oplus y) $.

Experimental results

Research questions

• RQ1 What is the exact combinatorial structure that determines the spectral norm of a symmetric Boolean function?
• RQ2 How does the spectral norm of a symmetric function relate to its decision tree complexity?
• RQ3 Can the spectral norm be tightly characterized using the minimal interval where the function becomes constant after parity shift?
• RQ4 Does the spectral norm characterization imply tight bounds on the communication complexity of symmetric xor functions?
• RQ5 To what extent does the spectral norm control the approximate monomial complexity of symmetric functions?

Key findings

• The logarithm of the spectral norm of a symmetric Boolean function is $ \Theta^*\left(r(f) \log(n / r(f))\right) $, where $ r(f) $ is the minimal interval size around the middle levels where the function or its parity-shifted version is constant.
• This characterization implies that the spectral norm tightly controls the complexity of parity decision trees, with the norm upper bounding the size of such trees.
• The result supports the Log Approximation Rank Conjecture for symmetric xor functions $ F(x,y) = f(x \oplus y) $, under Conjecture 5.1.
• It provides strong lower bounds on the approximate monomial complexity of symmetric functions with large $ r(f) $, via a connection to the Kushilevitz-Mansour learning algorithm.
• The characterization confirms that the spectral norm is a central complexity measure for symmetric functions, linking it to communication complexity and learning theory.
• The result offers a direct proof of the quantum-classical communication complexity equivalence for symmetric xor functions, as shown by Shi and Zhang, under the conjecture.