Computing local multipoint correlators using the numerical renormalization group

TL;DR

This paper introduces a numerical renormalization group method to accurately compute local multipoint correlators in strongly correlated quantum impurity systems across all temperature and frequency regimes, providing new insights into their dynamical properties.

Contribution

We develop a NRG-based approach to evaluate multipoint correlators using generalized spectral representations, enabling efficient and accurate calculations at low temperatures and frequencies.

Findings

Successfully computed three- and four-point correlators for the Anderson impurity model.

Resolved multiparticle excitations down to the lowest energies.

Applied method to RIXS spectra of impurity models.

Abstract

Local three- and four-point correlators yield important insight into strongly correlated systems and have many applications. However, the nonperturbative, accurate computation of multipoint correlators is challenging, particularly in the real-frequency domain for systems at low temperatures. In the accompanying paper, we introduce generalized spectral representations for multipoint correlators. Here, we develop a numerical renormalization group (NRG) approach, capable of efficiently evaluating these spectral representations, to compute local three- and four-point correlators of quantum impurity models. The key objects in our scheme are partial spectral functions, encoding the system's dynamical information. Their computation via NRG allows us to simultaneously resolve various multiparticle excitations down to the lowest energies. By subsequently convolving the partial spectral functions with appropriate kernels, we obtain multipoint correlators in the imaginary-frequency Matsubara, the real-frequency zero-temperature, and the real-frequency Keldysh formalisms. We present exemplary results for the connected four-point correlators of the Anderson impurity model, and for resonant inelastic x-ray scattering (RIXS) spectra of related impurity models. Our method can treat temperatures and frequencies -- imaginary or real -- of all magnitudes, from large to arbitrarily small ones.