The squared $L^2$-norm of the high-frequency part of f on a finite set S,
i.e. ∑_{x ∈ S} |f(x) − μ_S(f)|², where μ_S(f) = (1/|S|) ∑_{y ∈ S} f(y) is the average
of f over S. This is ‖f_h‖_{L^2(S)}² in the notation of BKT.
Instances For
Monotonicity of the high-frequency squared norm under enlarging the domain:
if S ⊆ T and S is nonempty, then ‖f_h‖²_{L^2(S)} ≤ ‖f_h‖²_{L^2(T)}. The proof
expands the mean of f on T versus the mean on S and uses that the cross-term
vanishes because ∑_{x ∈ S} (f(x) − μ_S) = 0.
Square-root version of monotonicity: ‖f_h‖_{L^2(S)} ≤ ‖f_h‖_{L^2(T)} whenever
S ⊆ T and S is nonempty.
BKT Lemma 3 (Subsection 7.3 of BKT): the high-frequency $L^2$-norm of f restricted to
the units ℤ_q^* is at most its high-frequency $L^2$-norm on all of ℤ_q, i.e.
‖f_h^*‖_{L^2(ℤ_q^*)} ≤ ‖f_h‖_{L^2(ℤ_q)}.