Deterministic replacement for the randomized subspace selection

Construct a deterministic procedure that, given a function $f$ and the dominant eigenspace of ${K}_{\varepsilon}(f\otimes\overline{f})$, finds a function $h$ satisfying the spectral separation and cardinality conditions required to recover quadratic characters via Theorem 7.1.

Background

The quadratic-character decomposition algorithm uses a random unit vector in the dominant eigenspace when the leading eigenvalues are not sufficiently separated. The authors prove that this randomized choice succeeds with high probability, but indicate that a deterministic choice should exist and leave its construction unresolved.

References

We believe that, instead of the random choice made in Theorem \ref{thm:main-random}, there is a deterministic procedure to find a function $h\in \textup{Eigen}\rho\big({K}\varepsilon(f\otimes \overline{f})\big)$ satisfying the properties required for Theorem \ref{thm:HiSpecBiject-intro} to be applicable. Investigating this lies outside the scope of this paper.

Spectral algorithms in higher-order Fourier analysis  (2501.12287 - Candela et al., 21 Jan 2025) in Remark following Algorithm "Quadratic character decomposition", Section 7