Unitary Invariants of the Finite Heisenberg Group

TL;DR

This paper demonstrates that unitary invariants of the finite Heisenberg group can separate generic orbits with much lower degree polynomials than polynomial invariants, revealing new insights into invariant theory and phase retrieval.

Contribution

It shows that degree-six unitary invariants suffice for orbit separation in the finite Heisenberg group, contrasting with higher degrees needed for polynomial invariants, and explores their applications.

Findings

Degree-six unitary invariants separate generic orbits.

Polynomial invariants require degree at least N.

Unitary invariants improve degree bounds in cyclic group representations.

Abstract

Polynomial invariants of a group action often appear only in high degree, and in many representations the invariant ring imposes severe degree constraints before any nontrivial invariants can occur. In contrast, the larger class of unitary invariants -- polynomials in both the variables and their conjugates -- typically exhibits very different behavior, and their separating power is comparatively unexplored. We highlight this contrast in the setting of the finite Heisenberg group $H_N$. Although the polynomial invariant ring $\mathbb{C}[V]^{H_N}$ contains no nontrivial elements below degree $N$, we show that degree-six unitary invariants are already sufficient to separate generic $H_N$-orbits up to a global phase factor. These invariants arise from cubic equations involving the magnitudes of a vector and its discrete Fourier transform. A single polynomial invariant in degree $N$ then resolves the remaining global phase, yielding full generic orbit separation. Our proof utilizes fundamental results from phase retrieval. Along the way we will also explore the utility of unitary invariants in obtaining improved degree bounds for representations of cyclic groups. This paper provides a concrete example in which the minimal separating degree for unitary invariants is dramatically lower than the minimal degree for polynomial invariants.