PaperlayerView exact revision on arXiv

Source-linked AI summary

A note on the convergence analysis of Laguerre approximations for analytic functions

Thomas Caussade, David P. Hewett

arXiv:2609.11259v2math.NAcs.NA

TL;DR

Wang claims root-exponential convergence for several Laguerre approximation problems, but this note argues that the proofs do not justify a limit under the integral sign. It proposes using dominated convergence, contingent on inequalities that remain conjectural, and gives rigorous estimates for a narrower meromorphic-function class.

  • Problem

    The proofs in do not properly justify the non-uniform limit under an integral sign needed for their claimed convergence results.

    Source: §1 · pp. 1–2

  • Method

    The note explains how dominated convergence could complete the proofs if certain nontrivial inequalities for generalized Laguerre polynomials and their weighted Cauchy transforms hold.

    Source: §5 · pp. 7–8

  • Results

    The inequalities needed for the general analytic-function results remain unproved, while rigorous estimates of the claimed form are derived for meromorphic functions with finitely many poles and algebraic behavior at infinity.

    Source: §5 · p. 8, §6 · p. 10

  • Takeaways & Limitations

    Proving the conjectured inequalities would complete the analysis of the results in.

    Source: §1 · p. 2, §5 · p. 8

    The inequalities needed to justify the dominated-convergence step have not been proved, and the relevant Cauchy-transform bound is left for future work.

    Source: §5 · pp. 8, 10

Abstract

from arXiv · show

In a recent paper (H. Wang, Math. Comp. 93, 2861-2884, 2024), Wang has presented a number of results concerning the convergence of approximations based on generalised Laguerre polynomials, claiming to have provided "the first rigorous proof of root-exponential convergence of Laguerre approximations for analytic functions". In this note we argue that the proofs of the main results of Wang's paper are incomplete, because they rely on taking a limit under an integral sign, and this is not properly justified. We explain how the proofs could be completed using the dominated convergence theorem, provided that certain nontrivial inequalities involving generalised Laguerre polynomials and their weighted Cauchy transforms hold. We make a related conjecture, which, if true, would complete the proofs of Wang's results. This conjecture remains unproven, but appears plausible, based on numerical experiments.

1 Introduction

Rigorous convergence theory for Laguerre approximations on the half-line is less developed than for analogous bounded-interval approximations. This note argues that Wang’s proofs of root-exponential convergence for analytic functions are incomplete and proposes a conjecture that could complete them.

1 Introduction

  • Wang claims root-exponential convergence for several Laguerre approximation and quadrature problems for functions analytic inside a parabola and algebraically bounded at infinity.The stated results cover coefficient and error estimates for projection and interpolation, spectral differentiation, and Gauss-Laguerre quadrature.Source: §1 · p. 1
  • The authors argue that proofs in take a non-uniform limit under an integral sign without adequately justifying it.They regard rigorous proof for the general analytic-function class considered in as an open problem.Source: §1 · p. 2
  • The note states a conjecture whose proof would complete the analysis in.The authors report that the conjecture remains unproven, though numerical experiments appear to support it.Source: §1 · p. 2

2 Notation and terminology

The note sets out notation for generalized Laguerre polynomials and their weighted Cauchy transform. It also gives an alternative direct proof of a Cauchy-transform representation used in Wang’s analysis.

2 Notation and terminology

  • For α > −1, P_n denotes polynomials of degree at most n, and Q_n comprises functions e^−x/2u(x) with u in P_n.The note uses α > −1 throughout.Source: §2 · p. 2
  • The generalized Laguerre polynomial L_n^(α) is expressed using the Pochhammer symbol and the confluent hypergeometric function M.The note also defines the weight ω_α(x) = x^αe^−x and the space L²(R+, ω_α).Source: §2 · p. 2
  • The weighted Cauchy transform L_n^(α) is analytic in C \ R+ and has a representation involving U(n + 1, 1 − α; −z).The note supplies an alternative, more direct proof of this representation, previously presented in [6, Lemma 2.1] without details.Source: §2 · p. 2
  • The parabola P_ρ is defined by Re[√−z] = ρ, and D_ρ is its interior containing R+.Fractional powers of complex variables use the principal branch.Source: §2 · p. 2

3 Main results of [6] and proof strategy

The section reviews the root-exponential bounds claimed in and the proof strategy used to establish them. It identifies an interpolation-error lemma issue and a more serious gap in justifying a limit under the integral sign.

3 Main results of [6] and proof strategy

  • The results in claim root-exponential bounds for quantities associated with Laguerre projection, interpolation, differentiation, and Gauss-Laguerre quadrature under a regularity condition on f.The bounds have the form Cn^γe^−c√n times an integral K, with constants independent of f and n.Source: §3 · p. 3
  • The proof strategy in establishes pointwise integrand convergence, bounds the integrands, passes the limit through the integral, and combines the resulting estimates.For the interpolation, differentiation, and quadrature results, Step 1 relies on [6, Lemma 4.1], which covered only β < 1/2; Lemma A.1 extends it to general β.Source: §3 · pp. 4–5
  • The more serious gap is that pointwise convergence of the integrands does not by itself justify convergence of the integrals in Step 3.The authors identify this as a nontrivial omission in the proofs of all the listed results in.Source: §3 · p. 5
  • Uniform relative-error convergence on P_ρ would suffice for the integral passage, but the authors say this condition does not hold for the listed cases.Thus that proposed justification cannot complete those proofs as presented.Source: §3 · p. 5

4 Pointwise convergence results

The pointwise convergence arguments in rely on large-n asymptotics for generalized Laguerre polynomials and their Cauchy transforms. These asymptotics do not supply the uniform control on the unbounded contour needed to justify the integral limit.

4 Pointwise convergence results

  • The pointwise limits used in rely on Perron’s formula for generalized Laguerre polynomials and large-n asymptotics for the Cauchy transform.The results cited in apply these asymptotics individually or in products, sometimes multiplied by e^z/2.Source: §4 · pp. 5–6
  • The asymptotics hold pointwise off R+ and uniformly on compact subsets, but not uniformly on unbounded subsets such as P_ρ.Their right-hand sides have exponential dependence on z, incompatible with the algebraic large-|z| behavior of the Laguerre polynomial.Source: §4 · pp. 6–7

5 Completing the analysis of [6]

The note gives sufficient conditions for using dominated convergence to complete Wang’s proofs, then conjectures inequalities that would establish those conditions. The inequalities remain unproved, though numerical experiments appear consistent with them.

5 Completing the analysis of [6]

  • If the stated inequalities hold, dominated convergence supplies a rigorous justification for the integral limits needed in the results from.The required condition is an integrable h(z) that bounds the relevant g_n(z) for all n ≥ n0.Source: §5 · p. 7
  • Theorem 5.1 gives sufficient conditions that would establish [6, Theorem 6.3] as well as Theorems 3.4, 4.2, and 4.3.When (24) and (25) hold, (26) follows automatically.Source: §5 · p. 7
  • For [6, Theorem 3.4], the assumed growth of f yields a |z|^-2 bound on g_n, giving an integrable dominator.The argument takes γ = β + 2 and sets h(z) = CC∞|z|^-2.Source: §5 · pp. 7–8
  • The proofs of [6, Theorems 4.2 and 4.3] follow similarly, but Theorem 4.3 requires the assumption on e^z/2f(z) to satisfy (8).The cited Perron formula’s uniform relative-error claim also conflicts with the polynomial nature of the Laguerre polynomials; a bounded-domain condition is presumed instead.Source: §5 · pp. 7–8
  • The authors conjecture that (24)–(26) hold for every ρ > 0 and α > −1, and that stronger bounds hold for every n0 ∈ N0.The stronger bounds imply (24)–(26) at specified n0 thresholds; the authors have not proved the inequalities for any ρ > 0 and α > −1.Source: §5 · p. 8
  • Plots near the apex of Pρ appear consistent with the conjectured bounds, but rigorous justification requires special-function estimates the authors have not obtained.The plotted cases vary ρ, α, n0, and n; the reported experiments focus on (27) and (28), which together imply (29).Source: §5 · pp. 8–10, Figure 2 · p. 9

6 Meromorphic functions

Rigorous convergence estimates of the stated form can be derived for meromorphic functions with finitely many poles and algebraic growth at infinity, by expressing the contour integral as a sum of residues. For a single-pole Gauss–Laguerre example, the note gives an error formula and discusses when the actual error may be small.

6 Meromorphic functions

  • For meromorphic functions with finitely many poles and algebraic behaviour at infinity, residue sums yield rigorous convergence estimates of the form (9).The note contrasts this result with the general-function estimates sought in.Source: §6 · p. 10
  • For a single simple pole, the Gauss–Laguerre quadrature error has the asymptotic rate |I − Q_n| = O(e^(−4Re[√−z0]√n) as n → ∞).The passage states that the contour can be deformed across the pole, leaving a residue contribution approximated using pointwise asymptotics.Source: §6 · p. 10
  • A similar formula for a pair of complex-conjugate simple poles was presented in [1, Equation (2.6)].The passage also relates it to [6, Remark 6.4].Source: §6 · p. 10
  • The contour plots show excellent agreement between the two sides of (38), especially for n + 1 = 30, provided z0 is not too close to the positive real axis.The quadrature formula Q_n blows up as z0 approaches a quadrature point.Source: §6 · p. 11
  • When Re[√−z0] is small, the actual quadrature error may still be acceptably small if Re[z0] is sufficiently large.The stated asymptotic rate may be slow in this regime; the error formula includes an exponential factor e^(−Re[z0]).Source: §6 · p. 11

A Appendix

The appendix provides proofs of auxiliary results related to the analysis in.

A Appendix

A.1 Proof of (4)

The appendix gives an alternative, more direct proof of (4) using the Rodrigues formula. It substitutes that formula into (2), integrates by parts, and extends the result by analytic continuation.

A.1 Proof of (4)

  • The proof of (4) uses the Rodrigues formula as an alternative to the earlier proof whose L’Hôpital-rule details were unclear.The Rodrigues formula is cited as [3, (18.5.5)].Source: §A.1 · p. 11
  • Substituting (39) into (2) and integrating by parts n times derives the representation for z ∈ C \ R+.There are no boundary terms because e^(−x)x^(n+α) and its derivatives through order n − 1 vanish at both endpoints when α > −1.Source: §A.1 · p. 11
  • For z ∈ (−∞, 0), a change of variable and the cited hypergeometric integral representation and connection formula establish (4), which analytic continuation extends to C \ [0, ∞).Source: §A.1 · pp. 11, 13

A.2 Extension of [6, Lemma 4.1]

Lemma A.1 extends the interpolation-error formula used in Step 1 of Wang’s proof strategy to general real β. Its proof applies Hermite’s interpolation remainder formula on a bounded contour and estimates the contour integrals under the condition n > β − 1.

A.2 Extension of [6, Lemma 4.1]

  • Lemma A.1 extends [6, Lemma 4.1], used in Step 1 for [6, Theorems 4.2, 4.3, and 6.3], from β < 1/2 to general β ∈ R.It assumes f is analytic on D_ρ, has growth |f(z)| = O(|z|^β) there at infinity, and n > β − 1.Source: §A.2 · p. 13
  • The lemma applies to interpolation at any n + 1 distinct positive nodes and bounds the error for x ∈ R+ using a contour enclosing x and all nodes.The proof uses Hermite’s formula for the interpolation remainder on the boundary formed by Γ and V.Source: §A.2 · p. 13
  • The contour estimates show that the integral over V tends to zero as η → ∞ when n > β − 1.The integral converges, and its lower integration limit tends to infinity with η.Source: §A.2 · p. 14

A.3 Estimating the constant K

Appendix A.3 estimates the constant K used in the bounds targeted by Results 3.1. Under a growth assumption on f, a lemma and an integral bound yield an upper bound for K.

A.3 Estimating the constant K

  • Results 3.1 seek to prove bounds of the form (9), which involve the constant K.Source: §A.3 · p. 14
  • If |f(z)| ≤ C|z|^β in Pρ, a lemma with a = 1 and d = b + β gives upper bounds for K in terms of exponentials and Γ functions.Source: §A.3 · p. 14
  • When d + 1/2 ≥ 0, the proof bounds (t^2 + ρ^2)^(d+1/2) by 2^(d+1/2)(t^(2d+1) + ρ^(2d+1)); the Γ-function integral representation then gives (51).The derivation uses parametrization (6).Source: §A.3 · p. 14
Loading 2609.11259v2…