Data bundle for the Euclidean projection bound theorem: a scale parameter R > 1,
a point set X of cardinality cardX, a direction set D of cardinality cardD,
a uniform projection bound S = max_{θ ∈ D} |π_θ(X)|, and local covering numbers
nX r, nD ρ measuring the maximum number of points of X (resp. directions of D)
in any ball of radius r (resp. ρ), each bounded above by cardX, cardD.
Instances For
Number of dyadic scales 1 ≤ r ≤ R used in the wave packet decomposition;
explicitly ⌊log₂ ⌈R⌉⌋ + 1.
Instances For
Existence of the wave packet decomposition f = ∑ₖ fₖ underlying the Euclidean
projection bound. Produces a total L²-energy together with per-scale energies and
pointwise values such that:
The total L²-energy of the wave packet decomposition extracted from
wave_packet_construction.
Instances For
The per-scale L²-energies ‖fₖ‖² extracted from wave_packet_construction.
Instances For
The pointwise values f(i) of the wave packet at the points of X,
extracted from wave_packet_construction.
Instances For
Bundles the defining properties of the wave packet construction: pointwise
lower bound f(i) ≥ |D|, the two energy identities, and the per-scale energy bound.
Pointwise lower bound for the wave packet: there exist values f(i) ≥ |D|
whose squared sum equals the total L²-energy.
Near-orthogonality of the dyadic pieces: the total L²-energy equals the
sum of the per-scale energies ∑ₖ ‖fₖ‖².
Lower bound on the wave packet L²-energy: since f(i) ≥ |D| at each of the
|X| points, we have ‖f‖² ≥ |D|² · |X|.
Trivial inequality form of orthogonality: ‖f‖² ≤ ∑ₖ ‖fₖ‖².
Per-scale L²-bound for the wave packet decomposition:
‖fₖ‖² ≤ S · |D| · R · max_{r ∈ [1,R]} (N_X(r) N_D(r/R) / r²).
The number of dyadic scales is at most log₂ R + 2, which produces the
logarithmic factor in the final Euclidean projection bound.
Upper bound on the wave packet L²-energy obtained by summing the per-scale
estimates over the log₂ R + 2 dyadic scales:
‖f‖² ≤ S · |D| · R · (log₂ R + 2) · max_{r ∈ [1,R]} (N_X(r) N_D(r/R) / r²).
Combined energy estimate chaining the lower and upper bounds:
|D|² · |X| ≤ S · |D| · R · (log₂ R + 2) · max_r (N_X(r) N_D(r/R) / r²).
Euclidean projection bound (the main theorem of the section): there is a constant
C > 0 such that
$$|D| \lessapprox \frac{S R}{|X|} \max_{1 \le r \le R} \frac{N_X(r)\, N_D(r/R)}{r^2}.$$
This is the Euclidean analogue of the Fourier projection bound (Theorem 2.3) and
follows from canceling a factor of |D| in the wave packet energy estimate.