PaperlayerView exact revision on arXiv

Source-linked AI summary

On a maximum principle in discretization with spherical harmonics

Arnold Reusken

arXiv:2610.13540v1math.NA

TL;DR

Approximating boundary data with spherical harmonics of degree at most N can make the discrete solution fail to satisfy the maximum principle. The paper studies the largest radius for which the maximum principle holds for the solution with approximated boundary data, across dimensions d≥2. For d=3, the radius bound holds for N ≥8 and is asymptotically sharp (for N →∞) apart from a factor 3.

  • Problem

    Approximating boundary data with spherical harmonics of degree at most N can make the discrete solution fail to satisfy the maximum principle.

    Source: §1 · p. 2

  • Method

    The paper studies the largest radius for which the maximum principle holds for the solution with approximated boundary data, across dimensions d≥2.

    Source: §1 · p. 2, §2 · p. 4, §5 · p. 9

  • Results

    For d=3, the radius bound holds for N ≥8 and is asymptotically sharp (for N →∞) apart from a factor 3.

    Source: §1 · p. 2

  • Takeaways & Limitations

    The analysis treats the discrete maximum principle as a fundamental property of spectral discretization and extends the radius question to arbitrary dimension d≥2.

    Source: §1 · p. 2

    For d≥3, the even-N approach used in the two-dimensional analysis does not apply, so the corresponding lemma is restricted to odd N.

    Source: §6 · pp. 16, 18

Abstract

from arXiv · show

Spherical harmonics have been studied extensively and are used in many fields of application. We consider the probably earliest application of spherical harmonics, namely for a representation of the solution of the Laplace equation on the $d$-dimensional unit sphere where the boundary data are represented in a spherical harmonics basis. We consider general $d \geq 2$. For the solution $u$ of the Laplace equation the fundamental maximum principle holds. If one approximates the boundary data in the finite dimensional space spanned by all spherical harmonics of degree at most $N$, the resulting discrete solution, denoted by $u_N$, does in general not satisfy the maximum principle. In this paper we derive sharp estimates for $r_N^{\rm max}$ such that the maximum principle holds for $u_N(x)$ for all $x \in \Bbb{R}^d$ with $\|x\| \leq r_N^{\rm max}$. We show that for $d=3$ one has $r_N^{\rm max} \gtrsim 1- 3 N^{-1} \ln\big( 2^\frac13 N \big) $. Similar estimates are derived for $d \neq 3$. We also study the sharpness of these estimates.

1. Introduction.

The introduction frames the loss of the maximum principle under finite spherical-harmonic approximation and studies the largest interior ball where it is retained. It extends the analysis to general dimensions and relates the result to positivity of a discrete Poisson kernel.

1. Introduction.

  • The paper treats the Laplace equation on spherical domains, representing boundary data with spherical harmonics and approximating the solution spectrally.The analysis applies for d≥2 and assumes essentially bounded boundary data that are continuous almost everywhere.Source: §1 · pp. 1, 3
  • Truncating the spherical-harmonic expansion can violate the continuous maximum principle, motivating a search for a smaller ball where it holds for uN.The introduction notes that high-frequency boundary errors are damped more strongly as 1−r increases.Source: §1 · p. 2
  • The analysis extends the discrete maximum-principle question from the three-dimensional case to arbitrary dimension d≥2.The paper identifies a two-dimensional precedent and presents its analysis for general dimensions.Source: §1 · p. 2
  • For d=3, the discrete maximum-principle radius bound holds for N≥8 and is asymptotically sharp apart from a factor of 3/2.The stated estimate also holds for small N, rather than only asymptotically.Source: §1 · p. 2
  • Positivity of the discrete Poisson kernel on a radius-bounded region is directly related to the maximum principle for the discrete solution.The main estimates follow from analyzing that kernel's positivity.Source: §1 · p. 3

2. Laplace equation and Poisson kernel in Rd.

This section represents the Laplace solution on the unit ball using the Poisson kernel and connects the maximum principle to properties of that kernel. It then motivates the discrete analysis by replacing the boundary data with a spherical-harmonic approximation.

2. Laplace equation and Poisson kernel in Rd.

  • The section considers the Laplace problem in the unit ball with essentially bounded boundary data that are continuous almost everywhere.The boundary condition is imposed almost everywhere on the sphere.Source: §2 · p. 3
  • The maximum principle for the Laplace problem is equivalent to nonnegativity of the Poisson kernel together with its unit-integral property.The kernel is nonnegative over r∈[0,1) and t∈[−1,1], and its integral over the sphere is 1.Source: §2 · p. 4
  • The discrete analysis replaces the boundary data f with a spherical-harmonic approximation fN and studies the resulting solution's maximum principle.Source: §2 · p. 4

3. Basics on spherical harmonics.

This section collects spherical-harmonic definitions and identities used to analyze the discrete Poisson kernel. It relates harmonics to homogeneous polynomials, Legendre polynomials, and orthogonal projections.

3. Basics on spherical harmonics.

  • Spherical harmonics are introduced as suitably scaled Laplace–Beltrami eigenfunctions, with homogeneous polynomials providing a more constructive alternative.The section gathers results needed for the discrete Poisson-kernel analysis.Source: §3 · p. 4
  • Legendre polynomials connect spherical-harmonic structure to the Poisson kernel through the Poisson identity.For d=2 and d=3, the section identifies them with Chebyshev and classical Legendre polynomials, respectively.Source: §3 · pp. 5–6
  • The section defines the space of spherical harmonics of degree at most N and its orthogonal projection.It also states an orthogonal decomposition and an alternative projection representation.Source: §3 · pp. 6–7

4. Spherical harmonics discretization and the discrete Poisson kernel.

The discretization replaces the boundary data f with its projection onto spherical harmonics of degree at most N. The resulting solution is described by a projected Poisson kernel, and the discrete maximum principle is generally guaranteed only on a restricted ball.

4. Spherical harmonics discretization and the discrete Poisson kernel.

  • The discrete solution uN is obtained by replacing the boundary data f with its projection PNf onto spherical harmonics of degree at most N.An orthonormal spherical harmonics basis is used to represent this projection.Source: §4 · p. 7
  • The discrete maximum principle generally does not hold throughout the unit ball, but the paper derives lower and upper estimates for the largest radius where it holds.The estimates describe how this radius depends on the spherical harmonics degree N.Source: §4 · p. 7
  • The projected kernel GN satisfies GN(r, ξ·η) = (PN Gr,ξ)(η) for r in [0,1) and ξ in Sd−1, and determines the dependence of uN on the original problem data f.Source: §4 · p. 8

5. Properties of the discrete Poisson kernel GN.

The section studies positivity of the discrete Poisson kernel to characterize where the discrete maximum principle holds. It establishes a unique maximal radius and relates its approach to 1 to bounds on its gap from 1.

5. Properties of the discrete Poisson kernel GN.

  • The discrete kernel GN retains the scaling property of the continuous Poisson kernel G.The argument uses GN=PNG.Source: §5 · p. 8
  • GN is not nonnegative for all radii in [0,1), so the discrete maximum principle does not extend to every radius in the unit ball.Figure 5.1 displays discrete kernels for N=7 and N=21.Source: §5 · p. 9, Figure 5.1 · p. 9
  • The section derives initial bounds for 1-rmax_N, then develops sharp upper and lower bounds that reveal how quickly this gap tends to zero with N.The sharp bounds are used to study how rmax_N tends to 1 as N varies.Source: §5 · pp. 9, 11
  • If GN(r0,t) is nonnegative for every t in [-1,1], it is nonnegative for every radius r in [0,r0].Positivity of the kernel at a radius provides the condition used to obtain the maximum principle on the corresponding ball.Source: §5 · pp. 9–10
  • There is a unique maximal radius rmax_N such that the discrete maximum principle holds on the ball of that radius, and rmax_N tends to 1 as N increases.For each fixed r<1, the discrete kernel GN(r,t) converges to G(r,t).Source: §5 · p. 11

6. Basics on Lambert W function.

The paper uses Lambert W estimates to derive bounds on the largest radius where the discrete maximum principle holds, then examines their sharpness across dimensions. The bounds capture logarithmic dependence on N in the cases discussed, with dimension- and parity-specific qualifications.

6. Basics on Lambert W function

  • The Lambert W function is introduced as the inverse of z ↦ ze^z and is used in deriving estimates involving r0(N).For N ≥ 4, r0(N) has no explicit expression in elementary functions, but the paper gives a lower estimate.Source: §6 · pp. 11–13

7. Bounds for 1 − rmax_N

  • For d = 2 and d = 3, lower bounds show that the ln N term in the upper bound cannot be avoided.The d = 3 lower-bound result is established for odd N.Source: §6 · pp. 16–17
  • For d = 3, the numerical results show distinct behavior for even and odd N, unlike d = 2.The lower-bound argument in d = 3 is restricted to odd N; the even-N approach relies on a Chebyshev-polynomial identity unavailable for d ≥ 3.Source: §6 · p. 18
  • For d = 2, the improved upper bound has the correct asymptotic behavior, while the first upper bound is asymptotically a factor of 2 too pessimistic.The paper relates the first bound's looseness to use of the triangle inequality.Source: §6 · p. 18
  • For d = 3, the upper bound has the correct asymptotic behavior apart from a factor of 3.The lower-bound analysis supports the logarithmic term's necessity for d = 3 and d = 2.Source: §6 · pp. 16, 18

8. Numerical experiments.

The numerical experiments estimate the maximum-principle radius by checking positivity of the discrete Poisson kernel over a discretized domain. The fitted curves closely match these estimates, and the d = 3 results distinguish even from odd N.

Numerical procedure

  • The numerical estimate δnum(N) is obtained by finding where the discretized minimum of GN(r, t) first becomes negative as r increases.The radius interval uses 101 points, while the t interval uses 20N equidistant points.Source: §8 · p. 18

Curve fitting

  • The fitted curve uses the form δ(a, b, N) = b ln(aN), with parameters chosen by weighted least squares.Measurements at larger N receive greater weight to reduce the influence of initialization effects.Source: §8 · p. 19

Numerical results

  • The fitted curves provide a very good fit to the numerical approximation of 1 − rmax_N, and d = 3 shows a difference between even and odd N.The experiments use N = 5k for k = 1, …, 40.Source: §8 · pp. 18–19

9. Appendix.

The appendix gives a proof of estimate (7.6) using the binomial series and includes a d = 2 numerical-results figure.

Appendix

  • The appendix proves estimate (7.6) using the binomial series.Source: §9 · pp. 19–20
  • For d = 2, Fig. 8.1 presents inverse numerical estimates, a least-squares fit, and two theoretical bounds.Source: Figure 8.1 · p. 20
  • The stated proof target in the appendix is inequality (7.6).Source: §9 · p. 20
Loading 2610.13540v1…