Abstract:Two Boolean functions f, g : F_2^{n} \to {-1, 1} are called linearly isomorphic if there exists an invertible matrix M \in F_2^{n\times n} such that f\circ M = g. Testing linear isomorphism is a generalization of, now classical in the context of property testing, isomorphism testing between Boolean functions. Linear-invariance of Boolean functions has also been extensively studied in other areas like coding theory and cryptography, and mathematics in general. In this paper, we will study the following two variants of this problem: [1] [Communication complexity: ] Assume that Boolean functions f and g on F_2^{n} are given to Alice and Bob respectively, and the goal is to test linear isomorphism between f and g by exchanging a minimum amount of communication, measured in bits, between Alice and Bob. Our main result is an efficient two-party communication protocol for testing linear isomorphism in terms of the approximate spectral norm of the functions. We will crucially exploit the connection between approximate spectral norm and sign-approximating polynomials. [2] [Query complexity: ] If f: F_2^{n} \to { -1, 1 } is a known function and g : F_2^{n} \to { -1, 1 } be an unknown function with a query access. We study the query complexity of testing linear isomorphism between f and g in terms of the approximate spectral norm of f. As in the case of communication complexity, we will use properties of the approximate spectral norm.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are two variants of the linear isomorphism test of Boolean functions: communication complexity and query complexity. Specifically: 1. **Communication Complexity**: Suppose two Boolean functions \( f \) and \( g \) are held by Alice and Bob respectively. The goal is to test whether these two functions are linearly isomorphic with the minimum amount of communication. The main result of the paper is to propose an efficient two - party communication protocol, and the communication complexity of this protocol is related to the approximate spectral norm of the function. 2. **Query Complexity**: Suppose a Boolean function \( f \) is known, while another Boolean function \( g \) is unknown but can be accessed by query. The research goal is to test whether these two functions are linearly isomorphic, and its query complexity is also related to the approximate spectral norm of the function. ### Main Contributions 1. **Communication Complexity Results**: - Theorem 1.1: For Boolean functions \( f, g: \mathbb{F}_2^n \to \{-1, 1\} \), let \( t=\min\{\|f\|_{1,1/3}, \|g\|_{1,1/3}\} \) and \( \epsilon, \omega > 0 \). The upper bound of the two - party communication complexity for deciding \( \delta_L(f, g) \leq \epsilon \) or \( \delta_L(f, g) \geq \epsilon+\omega \) is \( O(t^4 \log^2(1 / \omega)) \). 2. **Query Complexity Results**: - Theorem 1.3: For the known Boolean function \( g: \mathbb{F}_2^n \to \{-1, 1\} \), let \( t = \|g\|_{1, \alpha} \) and there is query access to the unknown function \( f: \mathbb{F}_2^n \to \{-1, 1\} \). The upper bound of the query complexity for deciding \( \delta_L(f, g) \leq \epsilon \) or \( \delta_L(f, g) \geq \epsilon+\omega \) is \( \tilde{O}((t / \omega)^{24}) \). ### Proof Ideas 1. **Proof Idea of Theorem 1.1**: - Determine which function (\( f \) or \( g \)) has a smaller 1/3 - approximate spectral norm. - Construct a Boolean function \( F \) such that \( \delta(f, F) \leq \omega / 4 \). - Reduce the amount of communication through linear transformation. - Use the linearly transformed function \( F \) to test the linear isomorphism distance. 2. **Proof Idea of Theorem 1.3**: - Use the Implicit Sieve Algorithm to estimate Fourier coefficients. - Construct a function \( \tilde{f} \) so that it has a high correlation with \( f \) under some unknown linear transformation. - Test the correlation between \( \tilde{f} \) and \( G \) under all linear transformations to decide \( \delta_L(f, g) \leq \epsilon \) or \( \delta_L(f, g) \geq \epsilon+\omega \). ### Conclusion By introducing the concept of approximate spectral norm, the paper significantly improves the efficiency of linear isomorphism testing. Especially for those Boolean functions whose approximate spectral norm is much smaller than the spectral norm, these results provide more efficient communication protocols and query algorithms.

Linear isomorphism testing of Boolean functions with small approximate spectral norm

Isoperimetric inequalities for real‐valued functions with applications to monotonicity testing

Is your function low-dimensional?

Boolean functions whose Fourier transform is concentrated on pairwise disjoint subsets of the input

A Dense Model Theorem for the Boolean Slice

On Fourier analysis of sparse Boolean functions over certain Abelian groups

Approximate polymorphisms

A Method for Determining the Affine Equivalence of Boolean Functions

Linear optics and the problem of Bell-like state discrimination

The Binary Linearization Complexity of Pseudo-Boolean Functions

Hypercontractivity, Sum-of-squares Proofs, and Their Applications

One-Way Communication Complexity of Partial XOR Functions

Computing Affine Equivalence Classes of Boolean Functions by Group Isomorphism.

An Average-Case Efficient Algorithm on Testing the Identity of Boolean Functions in Trace Representation.

High-dimensional general linear hypothesis tests via non-linear spectral shrinkage

Equivalence of concentration inequalities for linear and non-linear functions

Communication-constrained hypothesis testing: Optimality, robustness, and reverse data processing inequalities

Learning Polynomial Problems with $SL(2,\mathbb{R})$ Equivariance

Computing linear functions by linear coding over networks

Derandomized Non-Abelian Homomorphism Testing in Low Soundness Regime

Local Correction of Linear Functions over the Boolean Cube