Moment problems and bounds for matrix-valued smeared spectral functions
Ryan Abbott, William I. Jay, Patrick R. Oare · 2025
Read Moment problems and bounds for matrix-valued smeared spectral functions by Ryan Abbott, William I. Jay, Patrick R. Oare (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-01377.txt.
Introduction
Lattice physicists need curves that experiments cannot hand them directly. Ryan Abbott, William Jay, and Patrick Oare show a classic math problem can draw safe borders around the answer, in a 2025 paper. The curves are spectral functions, smeared and matrix-valued, reconstructed from correlation data computed on the lattice.
The authors prove the reconstruction tasks are equivalent to moment problems, a family studied deeply in mathematics. Theorems by Kovalishina then yield rigorous bounds on the smeared functions, and the paper implements those bounds numerically for the first time. The input, a positive-definite matrix of correlations, is exactly what variational spectroscopy studies already produce. The authors connect the method to recent Rayleigh-Ritz and Lanczos approaches and warn where finite computer precision bites. The result is a reconstruction with honest error fences.
Catalog record
| CREATOR | Ryan Abbott, William I. Jay, Patrick R. Oare |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-01377
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Moment problems and bounds for matrix-valued smeared spectral functions Preprint: MIT-CTP/5896 Ryan Abbott, William I. Jay, Patrick R. Oare Abstract Numerical analytic continuation arises frequently in lattice field theory, particularly in spectroscopy problems. This work shows the equivalence of common spectroscopic problems to certain classes of moment problems that have been studied thoroughly in the mathematical literature. Mathematical results due to Kovalishina enable rigorous bounds on smeared matrix-valued spectral functions, which are implemented numerically for the first time. The required input is a positive-definite matrix of Euclidean-time correlation functions; such matrices are routinely computed in variational spectrum studies using lattice quantum chromodynamics. This work connects the moment-problem perspective to recent developments using the Rayleigh–Ritz method and Lanczos algorithm. Possible limitations due to finite numerical precision are discussed. I Introduction The problem of numerical analytic continuation from a finite set of points arises frequently in lattice field theory. One example, and the focus of the present work, is reconstructing smeared spectral functions from Euclidean-time correlation functions. Another example appears in the finite-volume formalism for hadron spectroscopy, where one seeks to constrain scattering amplitudes in the complex energy plane given information at real energies Briceno et al. 2018 ; Salg et al. 2025 . The starting point for the present work is the Källén–Lehmann spectral representation of a Green’s function G(s) appearing in quantum field theory (QFT) Kallen 1952 ; Lehmann 1954 ; Schwartz 2014 , \displaystyle G(s)=\int_{0}^{\infty}ds^{\prime}\frac{\rho(s^{\prime})}{s^{\prime}-s+i\epsilon}, (1) where \rho is the spectral function satisfying \displaystyle\rho(s)=\frac{1}{\pi}\mathrm{Im}\,G(s)=\frac{1}{2\pi i}\discontinuity G(s). (2) Equation 1 makes manifest that evaluating the Green function at complex z=\omega+i\epsilon is equivalent to computing a smeared spectral function \rho_{\epsilon}(\omega) using a Poisson kernel \displaystyle\delta_{\epsilon}(\omega)=\frac{1}{\pi}\frac{\epsilon}{\omega^{2}-\epsilon^{2}}=\frac{1}{\pi}\mathrm{Im}\frac{-1}{\omega+i\epsilon}, (3) a property emphasized in Refs. Bergamaschi et al. 2023 ; Poggio et al. 1976 . Smeared spectral functions have appealing physical interpretations and are often accessible experimentally Davier et al. 2006 ; Davier et al. 2020 and through numerical calculation using lattice field theory Hansen et al. 2017 ; Bulava et al. 2022 . On the other hand, from a mathematical perspective, evaluating Eq. 1 for z\in\mathbb{C} is nothing more than the Stieltjes transform \displaystyle G(z)=\int_{\mathbb{R}}dx\,\frac{\rho(x)}{x-z}. (4) The Stieltjes transform clearly satisfies \displaystyle G(z)=G^{\dagger}(z^{*}), \displaystyle\frac{G(z)-G^{\dagger}(z)}{z-z^{*}}\succeq 0 (5) for z\in\mathbb{C}\setminus\mathbb{R} and the notation M\succeq 0 indicates that the matrix M is positive-definite. This transform arises naturally in the study of continued fractions, interpolation theory, and asymptotic series (see Ref. Kjeldsen 1993 for a historical overview). The present work focuses on the connection of the Stieltjes transform to moment problems, which are defined in Section II below. Connections between the Stieltjes transform and Padé approximations have been discussed previously in the context of lattice field theory Peris 2006 ; Aubin et al. 2012 . For instance, Ref. Aubin et al. 2012 used the Stieltjes transform to express the hadronic vacuum polarization using a convergent sequence of Padé approximants. Similar ideas have been explored recently in the condensed-matter literature. Ref. Nogaki et al. 2023 has discussed the Stieltjes transform and the connection to the Hamburger moment problem, focusing on a frequency-space conception of the problem. Ref. Fei et al. 2021a has considered analytic continuation of matrix-valued Green functions in frequency space. In contrast to Refs. Nogaki et al. 2023 ; Fei et al. 2021a , the present work connects moments of matrix-valued Green functions directly to the Euclidean-time correlation functions. The main results of the present work can be summarized as follows. First, the generic problem of determining a matrix-valued spectral function from a positive-definite matrix of Euclidean-time correlation functions can be viewed as a Hamburger moment problem. Second, existing results in the mathematical literature due to Kovalishina can be used to characterize the space of possible solutions to a given moment problem Kovalishina 1984 . To the authors’ knowledge, the present work gives the first numerical realization of Kovalishina’s analytic results. Third, these methods provide rigorous bounds on smeared matrix-valued spectral functions. Finally, the moment-problem perspective helps connects recent developments in the physics literature on spectral reconstructions, including the Rayleigh–Ritz method, the Lanczos algorithm, and Nevanlinna–Pick interpolation. The remainder of the work has the following structure. Section II provides a basic statement of moment problems. Section III makes the connection between moments problems and positive-definite matrices of Euclidean-time correlation functions calculable in lattice field theory. Section IV discusses Kovalishina’s general solution of the truncated Hamburger moment problem in terms of the Stieltjes transform Kovalishina 1984 , including rigorous bounds for individual components of the smeared matrix-valued spectral function. Section V provides some connections to recent ideas regarding the Rayleigh–Ritz method and the Lanczos algorithm. Section VI presents several numerical examples. Section VII discusses some practical considerations regarding input data with finite numerical precision. Section VIII closes with some conclusions. Appendix A collects various technical results. II Moment Problems The definition of a moment problem begins with the notion of a positive distribution, i.e., a matrix-valued function \rho_{ab}(x)\succeq 0 on the real line with indices a,b\in\{0,\dots,N-1\} , where N denotes the size of the matrix. 1 1 1 The mathematical literature on moment problems typically writes d\mu(x)\equiv\rho(x)\,dx and phrases theorems in the language of measure theory; that level of abstraction will not be required in the present work. The inequality is understood to mean that for each x\in\mathbb{R} , the matrix \rho_{ab}(x) is positive semidefinite. The scalar case corresponds to a familiar positive distribution \rho(x)\geq 0 on the real line. The indices a,b will frequently be suppressed when no confusion will arise. The n th moment with respect to the distribution is defined via \displaystyle C_{n}=\int dx\,x^{n}\rho(x), (6) where n is a non-negative integer. Provided the integrals converge, no conceptual complications arise in the “forward problem” of computing the moments given a distribution. The inverse problem—inferring an unknown distribution \rho(x) given the moments C_{n} —is known as a moment problem . Moment problems of particular interest for the present work are the Hausdorff moment problem ( \rho with support on the bounded interval [0,1] ), the Stieltjes moment problem ( \rho with support on the positive real line), and the Hamburger moment problem ( \rho with support anywhere on the real line). In practice, the focus of attention will be on the Hamburger moment problem, treating the others as special cases. A moment problem is referred to as truncated if only a finite subset of the moments is given. Since any numerical calculation is by necessity finite, all moment problems in the present work are understood to be truncated unless otherwise stated. Textbook treatments of moment problems include Refs. Akhiezer 2020 ; Schmüdgen 2017 . The connection between moment problems and the Stieltjes transform G(z) is well known Akhiezer 2020 ; Schmüdgen 2017 and can be seen from the asymptotic expansion \displaystyle G(z)=-\sum_{n=0}^{\infty}\frac{C_{n}}{z^{n+1}} (7) which follows immediately from Taylor expansion of Eq. 4 around z=\infty . In other words, the Stieltjes transform acts as a generating function for the moments of the distribution \rho . Eq. 2 , known in the mathematical literature as the Stieltjes–Perron inversion formula, then states that the Stieltjes transform encodes the same information as the original distribution. Given this equivalence, an alternative characterization of a moment problem is to recover G(z) from asymptotic data specified by the moments C_{n} . The asymptotic expansion