VC-dimension–degree trade-offs over finite abelian groups

Establish whether every non-zero Boolean function \(f:G\to\{0,1\}\) on a finite abelian group \(G\cong\prod_{j=1}^r\mathbb{Z}_{m_j}\) satisfies \({\rm VC}(f)+\deg(f)\ge {\rm rank}(G)\), where \({\rm VC}(f)\) is the VC-dimension of the family of left-translates of the support of \(f\), \(\deg(f)\) is the Fourier degree based on the number of nonzero frequency coordinates, and \({\rm rank}(G)=r\).

Background

The paper defines Fourier degree for functions on a finite abelian group by expressing the function in the character basis and measuring the largest number of nonzero coordinates in a frequency vector supporting a nonzero Fourier coefficient.

For a subset AGA\subseteq G, the relevant VC-dimension is the VC-dimension of the family of its left-translates, following the notation introduced by Rodgers and Sahay. The established inequalities for the Boolean cube motivate asking whether a comparable uncertainty principle holds for arbitrary finite abelian groups.

The proposed lower bound uses the minimal number of cyclic factors in a decomposition of GG as the ambient group dimension and specializes to the paper’s Boolean-cube inequality when G=Z2nG=\mathbb{Z}_2^n.

References

Can a similar trade-off relationship (VC-dimension v.s. degree) be established for Boolean functions over general finite abelian groups? i.e., does every non-zero Boolean function $f:G\to{0,1}$ satisfy

{\rm VC}(f) + \deg(f)\ge\; {\rm rank}(G)\ ?

where we define the ambient group dimension ${\rm rank}(G)$ as the minimal number of cyclic factors in the decomposition $G\cong\prod_{j=1}r\mathbb{Z}_{m_j}$ (this coincides with $n$ when $G=\mathbb{Z}_2n$).

VC-Dimension vs Degree: An Uncertainty Principle for Boolean Functions  (2510.13705 - Chang et al., 15 Oct 2025) in Section “Trade-off between VC-dimension and other Boolean function complexities beyond hypercube,” subsection on finite abelian groups