Constrained Curve Fitting
G. P. Lepage, B. Clark, C. T. H. Davies, K. Hornbostel, P. B. Mackenzie, C. Morningstar, H. Trottier
TL;DR
Constrained curve fitting addresses the underconstrained nature of lattice QCD analyses by injecting Bayesian priors into the fit of correlators modeled as $G_{ m th}(t)=\sum_{n=1}^{\infty} A_n e^{-E_n t}$. The method augments the traditional likelihood with a prior term to form ${\chi^2_{\rm aug}}$, allowing many higher-energy states to be included without destabilizing the leading parameter estimates and removing the need to optimally choose a $t_{\min}$. A full Bayesian treatment is presented, with Gaussian priors justified by maximum entropy and practical error estimation via quadratic approximations or bootstrap methods; the framework also supports empirical Bayes and maximum-entropy nonparametric approaches. Demonstrations across meson and heavy-quark correlators show precise extraction of low-lying energies while higher states are controlled by priors, enabling more accurate determinations of lattice spacings, quark masses, and broader extrapolations, with broad applicability to correlator analyses in lattice QCD and related fields.
Abstract
We survey techniques for constrained curve fitting, based upon Bayesian statistics, that offer significant advantages over conventional techniques used by lattice field theorists.
