PaperlayerView exact revision on arXiv

Source-linked AI summary

On the Gram matrix of standard inner products of asymmetrically-weighted Hermite functions

Ruiyang Dai

arXiv:2609.01144v1math.NAmath-ph

TL;DR

The paper addresses the structure and asymptotic behavior of Gram matrices arising from standard L2 inner products of scaled AW Hermite functions, motivated by stability questions in Vlasov–Poisson Galerkin methods. It derives explicit infinite-matrix formulas, factorizations, and inverses, then proves finite-matrix decay results including exponential decay of the smallest eigenvalue and Schur complement. These results support analysis and implementation of AW Hermite Galerkin spectral methods.

  • Problem

    Standard L2 products of AW Hermite functions generate Gram matrices relevant to Galerkin Vlasov–Poisson methods, whose finite forms are dense and ill-conditioned and whose asymptotic properties require analysis.

  • Method

    The paper derives explicit entries, a Cholesky factorization, and an inverse for the infinite Gram matrix, interprets the factorization using a scaled Bargmann–Fock basis, and analyzes finite matrices and their Schur complements.

  • Results

    The smallest eigenvalue of the finite Gram matrix decays exponentially to zero, with exponential rate 3−N, and the Schur complement also exhibits exponential decay.

  • Takeaways & Limitations

    The derived formulas and asymptotic properties can be used in the numerical analysis and implementation of Galerkin spectral methods for the Vlasov–Poisson system.

  • Takeaways & Limitations

    The Bargmann–Fock factorization identity is an identity of sesquilinear forms on finitely supported sequences, not a bounded operator identity on ℓ2 because D−1 grows exponentially.

Abstract

from arXiv · show

Let A denote the infinite Gram matrix associated with the standard L2 inner product of asymmetrically-weighted (AW) Hermite functions. We derive an explicit representation of its entries and its Cholesky factorization. We further show that this factorization admits a natural interpretation on a scaled Bargmann-Fock basis. An explicit formula for the inverse of A is also obtained. We then consider the corresponding finite Gram matrix and analyze its asymptotic property, as well as that of its Schur complement. The analysis is motivated by numerical methods for plasma physics, in particular Galerkin spectral methods applied to the Vlasov-Poisson (VP) system. As an application, we demonstrate how the derived Gram matrix formulas and asymptotic results can be exploited in the analysis and implementation of a Galerkin spectral method for the VP system.

1 Introduction

The paper studies Gram matrices generated by scaled asymmetrically weighted Hermite functions, motivated by numerical stability in Galerkin discretizations of the Vlasov–Poisson system. It develops explicit matrix formulas and analyzes finite-dimensional conditioning and asymptotics.

  • Motivation: Scaled AW Hermite functions combine Hermite polynomials with Gaussian decay and can reduce the number of modes needed for approximately Gaussian profiles.They are widely used for numerical discretization in physics and are suited to plasma problems with exponential velocity decay.
  • Gram matrices: The standard L2 inner product of scaled AW Hermite functions produces an infinite Gram matrix and finite matrices formed from the first N + 1 functions.The finite matrix is dense and ill-conditioned, motivating analysis of its asymptotic properties.
  • Contributions: The finite Gram matrix and its Schur complement are analyzed asymptotically, including the decay of the smallest eigenvalue.The study is connected to numerical analysis and implementation of Galerkin spectral methods for the Vlasov–Poisson system.
  • Application context: AW Hermite Petrov–Galerkin formulations are widely used for the Vlasov–Poisson system but are known to suffer from numerical instability.The Gram matrix appears because AW Hermite functions are not orthogonal under the Galerkin inner product.
  • Contributions: The paper derives explicit entries, a Cholesky factorization, and an inverse for the infinite Gram matrix, with a scaled Bargmann–Fock interpretation of the factorization.These results are presented as tools for understanding the matrix structure and its use in spectral methods.

2 Hermite functions

The paper introduces weighted-function notation and recursive constructions for scaled AW Hermite functions, while distinguishing their weighted orthogonality from the usual L2(R) setting. Parity further implies that mixed even–odd inner products vanish.

  • Weighted setting: The weighted Lebesgue space consists of measurable functions on R satisfying the paper’s weight-based integrability condition.The associated weighted inner product is used to formulate the AW Hermite system.
  • Hermite functions: Scaled AW Hermite functions form an orthogonal system under the weighted inner product and can be defined recursively.The paper gives both a recursive scheme and an additional recurrence relation for successive functions.
  • Inner products: The analysis also considers the usual Lebesgue space L2(R), where the scaled AW Hermite functions are generally not orthogonal under the standard inner product.This distinction leads to the Gram matrix studied later in the paper.
  • Parity: Because ψ_m is even for even m and odd for odd m, the integral of ψ_m(v)ψ_n(v) over R vanishes for opposite parities.This parity property simplifies the structure of standard inner products.

3 Infinite Gram matrix

The paper derives explicit formulas for the infinite Gram matrix of scaled AW Hermite functions, including its entries, Cholesky factorization, Bargmann–Fock interpretation, and inverse decomposition.

  • 3.1 Coefficients of the infinite Gram matrix: The Gram-matrix coefficients are finite scalar products because products of two AW Hermite functions equal a Gaussian multiplied by a polynomial.The paper proposes formulas because exact coefficient values were not available in the cited special-functions literature.
  • 3.1 Coefficients of the infinite Gram matrix: If m+n is odd, the Gram-matrix entry a_mn vanishes because the integrand is a Gaussian multiplied by an odd polynomial.For even m+n, the paper derives an explicit expression using Hermite-polynomial identities.
  • 3.1 Coefficients of the infinite Gram matrix: The coefficient formulas admit computationally stable recurrences using only multiplication by positive numbers no greater than one.These recurrences avoid unstable direct calculations involving factorials of large natural numbers.
  • 3.2 Cholesky decomposition of the infinite Gram matrix: The infinite Gram matrix A has a unique Cholesky decomposition A=LDL^T with a unit lower-triangular L and diagonal D.The factorization follows from positive definiteness and explicit formulas for the entries of L and D.
  • 3.3 Bargmann–Fock space interpretation: On a scaled Bargmann–Fock basis, the factorization is represented through a rescaled transform whose normalization 2^-1/4 is unitary from L2(R) onto F2(C).The matrix L represents the action of an exponential differential operator on the normalized monomial basis.
  • 3.4 Inverse of the infinite Gram matrix: The inverse Gram matrix admits a decomposition involving the Gram kernel matrix Z and diagonal matrix D, but the identity is only asserted as a quadratic-form identity on finitely supported sequences.It is not a bounded operator identity on ℓ2 because D^-1 grows exponentially.

4 Finite Gram matrix

The finite Gram matrix is studied through block decompositions inherited from the infinite matrix, yielding Gram-kernel relations and asymptotic bounds for off-diagonal blocks and Schur-complement terms.

  • 4 Finite Gram matrix: The infinite matrices A, Z, L, and D are decomposed into blocks at a fixed truncation index N to relate the infinite and finite Gram matrices.The finite matrix consists of the first N+1 scaled AW Hermite functions.
  • 4 Finite Gram matrix: The block matrices satisfy A11Z12+A12Z22=0, linking the finite Gram-matrix blocks with the Gram kernel.The relation follows from the Cholesky factorization and the corresponding block identities.
  • 4 Finite Gram matrix: The Gram kernel matrix Z is named because each of its columns belongs to the kernel of the truncated block Gram matrix.This kernel structure is used in the block analysis of A and Z.
  • 4 Finite Gram matrix: The entries of A21A11^-1 are bounded through explicit estimates derived from the Cholesky decomposition and block relations.The paper also gives an upper bound for the entries of the corresponding block expression.
  • 4 Finite Gram matrix: For indices n,m≥N+1, the diagonal entries of A dominate the absolute values of entries in their rows, while diagonal magnitudes decrease with row index.The resulting asymptotic behavior is consistent with the O(N^-1/4) amplitude of scaled AW Hermite functions in their main support.

5 Decay properties of the finite Gram matrix

The finite Gram matrix is analyzed through its smallest eigenvalue and Schur complement, with explicit asymptotic formulas and decay estimates. The smallest eigenvalue decays exponentially, while the Schur-complement estimate is sharper.

  • 5.1 Smallest eigenvalue: The smallest eigenvalue λN is estimated through the Rayleigh quotient of the symmetric positive-definite finite Gram matrix.The argument uses positivity of the inverse and bounds its largest eigenvalue through matrix entries.
  • 5.1 Smallest eigenvalue: The Rayleigh-quotient lower bound is shown to be asymptotically sharp by an appropriate choice of coefficients xn.The proof follows the stated coefficient construction and asymptotic estimates for large N.
  • 5.1 Smallest eigenvalue: The smallest eigenvalue decays with exponential rate 3−N as N tends to infinity.The result follows by inverting the asymptotic relation obtained for the relevant maximum.
  • 5.2 Schur complement: The finite Gram matrix is partitioned into blocks to obtain an explicit formula for its Schur complement.The block dimensions and scalar corner entry are specified before the explicit complement is derived.

6 Conclusion

The paper derives explicit structural and asymptotic results for Gram matrices of AW Hermite functions, including consequences for Galerkin spectral methods for the VP system. These results also connect to broader questions about orthogonal polynomials, Gram matrices, and approximation theory.

  • 6 Conclusion: The work derives explicit coefficients, a Cholesky decomposition, and an inverse formula for the infinite Gram matrix of AW Hermite functions.The Cholesky factorization is additionally interpreted using a scaled Bargmann–Fock basis.
  • 6 Conclusion: The finite Gram matrix has decay properties, including exponential decay of its smallest eigenvalue and of its Schur complement.These properties are presented as useful for numerical analysis of AW Hermite Galerkin methods applied to the VP system.
  • 6 Conclusion: The derived formulas and asymptotic results may support analysis and implementation of Galerkin spectral methods for the Vlasov–Poisson system.The paper frames this as a numerical-analysis application of the Gram-matrix results.
  • 6 Conclusion: The Gram-matrix results may also enrich orthogonal-polynomial and Gram-matrix theory and have broader implications for approximation theory.The stated broader implications extend beyond the direct numerical-analysis applications.

A.1 Trace estimate

The trace estimate is obtained by separating even- and odd-indexed diagonal contributions and analyzing their large-index behavior. Stirling’s formula and interior Laplace’s method identify the leading asymptotic contribution.

  • A.1 Trace estimate: For even N = 2M, the proof separates the even-indexed diagonal entries from the remaining contribution.The even contribution is indexed by r = 0, . . . , M and is then reorganized for asymptotic analysis.
  • A.1 Trace estimate: The even contribution is reparameterized with k = m − r and analyzed using a continuous variable x = k/m.This converts the combinatorial sum into a form suitable for asymptotic estimation.
  • A.1 Trace estimate: Stirling’s formula approximates the factorial terms appearing in the diagonal-entry sums.The resulting asymptotics are combined with an integral approximation.
  • A.1 Trace estimate: Interior Laplace’s method determines the leading behavior by locating the interior maximum of the associated continuous function.The method replaces the sum by an integral and evaluates its dominant contribution.
  • A.1 Trace estimate: The odd-indexed contribution is treated separately, after which the even and odd contributions are collected to obtain the trace estimate.The final collection gives the asymptotic diagonal sum for large indices.

A.2 Asymptotic behavior in the bulk region

The bulk-region analysis examines inverse Gram-matrix entries whose indices lie within O(N^1/4) of the matrix boundary. The leading contribution is asymptotically determined by the terminal index k = N.

  • A.2 Asymptotic behavior in the bulk region: For fixed α > 0, the bulk region consists of indices satisfying N − αN^1/4 ≤ i, j ≤ N.Equivalently, i = N − a and j = N − b with a, b = O(N^1/4).
  • A.2 Asymptotic behavior in the bulk region: The analysis studies the diagonal entries of A11 and A11−1 together with the Gram-kernel entries in this bulk regime.These quantities are compared through the asymptotic expressions indexed by i and j.
  • A.2 Asymptotic behavior in the bulk region: Writing k = N − 2r leaves only O(N^1/4) admissible values of r in the bulk-region sum.Stirling’s formula is then applied to the resulting factorial expressions.
  • A.2 Asymptotic behavior in the bulk region: The asymptotic factors for the offsets a and b are combined through the coefficients Cr(a, b)N−2r.The leading term is isolated after expanding the relevant factors.
  • A.2 Asymptotic behavior in the bulk region: The contribution k = N asymptotically determines all three quantities considered in the bulk analysis.This identifies the terminal summation index as the dominant contribution.

B Application to the VP system

The paper applies AW Hermite functions as trial and test functions in a Galerkin spectral method for the one-dimensional Vlasov–Poisson system. Because these functions are not orthogonal under the standard L2 inner product, the resulting formulation involves Gram-matrix blocks.

  • B Application to the VP system: The method uses AW Hermite functions as both trial and test functions for the Vlasov–Poisson system.The formulation is developed in velocity space with periodic spatial boundary conditions and a discretized Poisson equation.
  • B Application to the VP system: Substituting the spectral expansion and testing the Vlasov equation with ψn produces the semi-discrete system coupled to the discretized Poisson equation.The derivation uses the recurrence relations of the AW Hermite functions.
  • B Application to the VP system: Non-orthogonality in the standard L2(R) inner product introduces the Gram-matrix blocks A11 and A12 into the Galerkin method.This distinguishes the formulation from the Petrov–Galerkin method.
  • B Application to the VP system: The resulting formulation provides a Galerkin spectral method for the Vlasov–Poisson system in velocity space.The subsequent treatment addresses both the Gram matrix and conservation properties.

B.1 Sparse structure

Although the Galerkin formulation contains a dense Gram matrix, an equivalent system preserves the sparse structure of the classical Petrov–Galerkin method with only a small structural modification.

  • B.1 Sparse structure: Direct use of the dense Gram matrix can incur substantial computational costs.The paper therefore reformulates the semi-discrete system rather than applying the dense matrix directly.
  • B.1 Sparse structure: The reformulated system modifies the transport or acceleration matrix only through entries in its last column.This follows because B21 or D21 has a single nonzero entry in its first row and last column.
  • B.1 Sparse structure: The equivalent formulation preserves the sparsity structure of the classical Petrov–Galerkin method.The additional last-column entries are the only stated departure from that structure.
  • B.1 Sparse structure: The reformulation avoids direct use of the dense Gram matrix while introducing only negligible additional computational cost.The sparse implementation is obtained using the factorization identity invoked through Lemma 4.4.

B.2 Sharper estimate

The paper compares two approaches for estimating the highest moment in the spectral method: a bound based on the smallest Gram-matrix eigenvalue and one based on the Schur complement. The Schur-complement estimate is sharper.

  • B.2 Sharper estimate: A sharp estimate for the highest moment u_N is particularly relevant for spectral methods applied to the Vlasov–Poisson system.The analysis assumes a condition involving a function g(N) that tends to zero as N tends to infinity.
  • B.2 Sharper estimate: The smallest-eigenvalue approach estimates the relevant quantity using Lemma 5.1 and the Cauchy–Schwarz inequality.The paper presents this as one route to bounding the highest moment.
  • B.2 Sharper estimate: The alternative approach derives an estimate from the Schur complement.This estimate is presented through Lemma 5.2.
  • B.2 Sharper estimate: The Schur-complement estimate is sharper than the estimate based on the smallest eigenvalue of the Gram matrix.The comparison is the section’s stated conclusion.
Loading 2609.01144v1…