Source-linked AI summary
Positivity loss in bandlimited spectral reproduction on spheres
Hao-Ning Wu
TL;DR
The paper asks how closely an N-bandlimited operator can combine exact reproduction through degree L with positivity. It proves a sharp quadratic positivity-loss bound and obtains it by correcting a positive Jackson operator with a smooth filter, with consequences for Fourier summability, filtered hyperinterpolation, and maximum principles.
Problem
Spectral operators should exactly reproduce low-frequency modes while remaining finitely bandlimited, but these requirements conflict with positivity.
Method
The paper proves a lower bound using a Fejér peak test and concentration estimates, then matches it by correcting a positive Jackson operator with a smooth filter.
Results
The smallest positivity loss has sharp order (L/(N + 1))^2 for 1 ≤ L < N, while positivity, exact reproduction, and finite bandwidth cannot all hold simultaneously.
Takeaways & Limitations
The quadratic obstruction quantifies the transition-band trade-off and extends to classical Fourier summability, filtered hyperinterpolation, and maximum-principle questions.
Takeaways & Limitations
For filtered hyperinterpolation, the reproduction claim assumes a cubature rule exact to degree at least L + N_L.
Abstract
from arXiv · showhide
How small can the positivity loss be for an $N$-bandlimited spectral operator that exactly reproduces all modes up to degree $L$? For spherical polynomial approximation on $\mathbb S^d$, we prove that the smallest possible excess of the uniform operator norm above 1, equivalently the least positivity loss, is of sharp order $\left({L}/{(N+1)}\right)^2$ when $1\le L< N$. The lower bound follows from a Fejér peak test and a concentration estimate for bandlimited kernels, while a matching upper bound is obtained by correcting a positive Jackson operator with a smooth filter. We illustrate the result in three settings. On the circle, taking $N=sL-1$, this determines the sharp order of the generalized-projection constant above 1 and identifies the gap between the $s^{-1}$ excess of delayed de la Vallée--Poussin means and the optimal $s^{-2}$ order. For filtered hyperinterpolation, whose operator norm has long been known to be uniformly bounded, we give a quantitative lower bound on its separation from the positivity threshold 1. Finally, we identify an operator-level obstruction to maximum principles.
1 Introduction
The paper studies operators that exactly reproduce spherical polynomials through degree L while remaining bandlimited to degree N, focusing on the unavoidable positivity loss caused by imposing both constraints. It defines the associated generalized-projection constant and proves that the optimal excess norm above 1 has sharp quadratic order in L/(N+1).
- Problem: Exact reproduction through degree L and output bandwidth N create a transition band of degrees L+1 through N, but positivity is impossible when L≥1 and N is finite.Without exact reproduction or finite output bandwidth, positive operators are available; together, the constraints force nonzero positivity loss.
- Problem: The two-scale generalized-projection constant measures the smallest uniform operator norm among operators reproducing P_L with range contained in P_N.The condition A|P_L=I is the exact reproduction property, and the extremal norm is compared with the positivity threshold 1.
- Positivity loss: The positivity loss equals half the excess of the operator norm above 1, so minimizing one quantity is equivalent to minimizing the other.The identity also establishes that 1 is the smallest possible norm and that it is attained exactly by positive operators.
- Main result: The smallest positivity loss has sharp order (L/(N+1))^2, showing that a wider transition band approaches positivity only at a quadratic rate.Theorem 1.3 provides dimension-dependent constants for this order when the two spectral scales satisfy the stated regime.
- Applications: On the circle with N=sL−1, delayed de la Vallée–Poussin means have s−1 excess norm, whereas the optimal admissible order is s−2.The result identifies a strictly smaller optimal order over the full admissible operator class.
- Proof strategy: The proof combines a Fejér peak and bandlimited-kernel concentration for the lower bound with a positive Jackson operator and smooth-filter correction for the matching upper bound.The paper also reduces the extremal problem to zonal operators and relates their positivity loss to kernel negative mass.
2 Preliminaries
The preliminaries establish the spherical harmonic and zonal-kernel framework used to represent bandlimited operators. They show how spectral multipliers reproduce low-degree harmonics and how positivity loss becomes the negative mass of a zonal kernel.
- Spherical harmonic framework: Spherical harmonics H_ℓ provide the orthogonal frequency decomposition of functions on S^d, while P_n consists of spherical polynomials of degree at most n.The spaces H_ℓ have dimensions Z(d,ℓ), and P_n is built from harmonics through degree n.
- Zonal operators: Zonal kernels have the form K(x,y)=k(x·y), and their integral operators are diagonal with respect to the spherical harmonic decomposition.The Funk–Hecke formula supplies the corresponding spectral action.
- Spectral constraints: A polynomial kernel of degree at most N produces an operator with bandwidth at most N, and exact reproduction of P_L requires spectral coefficients a_ℓ=1 for 0≤ℓ≤L.For a zonal operator, unitality is equivalent to a_0=1.
- Positivity loss: For a unital zonal operator, positivity loss is represented by the negative mass of its kernel.Rotation invariance makes the relevant integral independent of the evaluation point.
- Positivity loss: For zonal operators, kernel negative mass equals positivity loss and half the excess norm, and it is also the worst violation of the interval [0,1].This connects the kernel representation directly to the operator-norm formulation.
3 Symmetry reduction and the lower bound
The lower-bound argument reduces arbitrary admissible operators to zonal kernels without increasing norm, then combines a Fejér peak test with bandlimited concentration to force positivity loss.
- Symmetry reduction: Averaging over O(d + 1) preserves reproduction and bandwidth while not increasing the operator norm, reducing the extremal problem to zonal operators.The averaged operator is O(d + 1)-equivariant and has a zonal kernel.
- Symmetry reduction: The zonal formulation minimizes positivity loss over kernels of bandwidth at most N that reproduce PL exactly.This yields the equivalent operator-norm and positivity-loss extremal problem.
- Positivity loss: 1 + 2 inf p(TK) equals the infimum operator norm, identifying positivity loss as the exact excess above the threshold 1.The identity follows from the representing-kernel formulation and unitality.
- Lower-bound mechanism: A low-frequency polynomial qL,e vanishes at e while growing away from e at scale L^-1, providing the Fejér peak test.The construction satisfies 0 ≤ qL,e ≤ 1 and qL,e(e) = 0.
- Lower-bound mechanism: A bandwidth-N kernel with normalized integral cannot concentrate all positive mass inside a spherical cap of radius comparable to (N + 1)^-1.The concentration estimate supplies the spatial scale needed to test the kernel outside the peak.
- Lower-bound conclusion: Theorem 3.5 combines the peak and concentration estimates to give a uniform lower bound for every admissible operator when 1 ≤ L ≤ N.The resulting bound is the lower half of the sharp positivity-loss estimate.
4 The matching upper bound
The matching upper bound corrects a positive Jackson approximation with a smooth flat-top filter, restoring exact reproduction while retaining quadratic approximation error.
- Positive Jackson approximation: A positive Jackson operator JN has range in PN, preserves constants and nonnegativity, and has operator norm 1.Its error on polynomials of degree m is of order (m/(N + 1))^2.
- Positive Jackson approximation: The Jackson kernel is normalized and nonnegative, with degree at most s(k − 1), making the associated operator positive and bandlimited.The construction chooses its normalization so that constants are reproduced.
- Flat-top correction: A smooth filter hβ produces a uniformly bounded filtered operator VL that reproduces PL and has range within the available transition bandwidth.Its support conditions give exact reproduction and finite spectral range.
- Flat-top correction: The corrected operator TL,N = JN + (I − JN)VL combines positivity with a flat-top correction that restores exact reproduction.The correction acts on VLf, which lies in a low-degree polynomial space where Jackson approximation is controlled.
- Upper-bound conclusion: For every fixed 0 < κ < 1 and 1 ≤ L ≤ κN, the constructed zonal operator belongs to the admissible class and satisfies the matching upper bound.Together with the lower bound, this proves the sharp order claimed in the paper.
5 Applications and discussion
The applications show how the quadratic positivity-loss bound governs generalized projections, filtered hyperinterpolation, and maximum-principle violations. Across these settings, transition bands improve stability but cannot eliminate the separation from positivity.
- Applications and discussion: Fejér means preserve positivity and finite bandwidth but reproduce only constants, whereas delayed de la Vallée–Poussin means restore exact low-frequency reproduction at the cost of positivity.This illustrates the paper’s impossibility triangle for positivity, exact reproduction, and finite bandwidth.
- 5.2 Filtered hyperinterpolation: Filtered hyperinterpolation retains exact reproduction of PL while smooth filtering provides uniform stability, yet its norm remains quantitatively separated from the positivity threshold 1.The lower bound depends only on the ratio of reproduced to output bandwidths, not on the transition profile or cubature rule.
- 5.2 Filtered hyperinterpolation: The filtered-hyperinterpolation lower bound applies under cubature exactness of degree at least L+NL, while the associated filtered kernel is built from a smooth filter.The filter equals 1 through degree L and vanishes from degree aL onward, as specified in the construction.
- Maximum-principle obstruction: Exact low-frequency reproduction and finite bandwidth preclude a maximum principle on [0,1], with the smallest possible worst-case violation having quadratic order.The obstruction is static and operator-level, unlike effective maximum principles for evolution discretizations that control solutions in enlarged invariant intervals.