Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2007.13610

Analytic continuation of the self-energy via Machine Learning techniques

Taegeun Song, Roser Valentí, Hunpyo Lee

Published Jul 27, 2020• 3 citations
arXivHTMLPDF

Abstract

We develop a novel analytic continuation method for self-energies on the Matsubara domain as computed by quantum Monte Carlo simulations within dynamical mean field theory (QMC+DMFT). Unlike a maximum entropy (maxEn) procedure employed for the last thirty years, our approach is based on a machine learning (ML) technique in combination with the iterative perturbative theory impurity solver of the dynamical mean field theory self-consistent process (IPT+DMFT). The input and output training datasets for ML are simultaneously obtained from IPT+DMFT calculations on Matsubara and real frequency domains, respectively. The QMC+DMFT self-energy on real frequencies is determined from the -- usually noisy -- input QMC+DMFT self-energy on the Matsubara domain and the trained ML kernel. Our approach is free from both, bias of ML training datasets and from fitting parameters present in the maxEn method. We demonstrate the efficiency of the method on the testbed frustrated Hubbard model on the square lattice.

DOI
10.48550/arxiv.2007.13610

Related papers

Analytic Continuation of Quantum Monte Carlo Data: Optimal Stochastic Regularization Approach

Igor S. Krivenko, Alexey N. Rubtsov

A new algorithm for analytic continuation of noisy quantum Monte Carlo (QMC) data from the Matsubara domain to real frequencies is proposed. Unlike the widely used maximum-entropy (MaxEnt) procedure, our method is linear with respect to input data and can therefore be applied to off-diagonal components of a thermal Green's function, or to a self-energy function. The latter possibility is used to analyze QMC results for the half-filled single-band Hubbard model on a Bethe lattice at a low temperature. Our method qualitatively resolves peaks near the inner edges of the Hubbard bands in the vicinity of a Mott transition, whereas a MaxEnt procedure does not. An existence of such structures has been clearly established before in a high-precision D-DMRG calculation by Karski et al. We also analyze a stability of the new method subject to changes of adjustable parameters.

arXiv:cond-mat/0612233•Dec 9, 2006•4 citations

Algorithm for computing perturbation series of dynamical mean field theory

Corentin Bertrand, Michel Ferrero, Olivier Parcollet

We show how to use diagrammatic techniques to compute the weak-coupling perturbation series of the self-consistent solution to a dynamical mean field theory (DMFT) problem. This approach constitutes an alternative to using diagrammatic techniques directly as an impurity solver. It allows one to bypass the need of multiple perturbative series resummations within the DMFT self-consistency loop. It can be applied at or out of equilibrium, with any diagrammatic formalism, such as real times, imaginary times, or Matsubara frequencies formalisms. As a proof of principle, we illustrate our method with the half-filled Hubbard model on the Bethe lattice in the DMFT approximation, using quantum quasi Monte Carlo (QQMC) to obtain the impurity perturbation series on the real time axis.

arXiv:2410.00497•Physical review. B./Physical review. B•Apr 14, 2025•0 citations

A Mathematical Analysis of IPT-DMFT

Eric Cancès B, Alfred Kirsch, Solal Perrin-Roussel

Abstract We provide a mathematical analysis of the Dynamical Mean-Field Theory (DMFT), a celebrated representative of a class of approximations in quantum mechanics known as embedding methods . Within the simplest setting of this approximation, which we introduce briefly at this point, we prove that, under certain assumptions, the DMFT equations admit a solution for any set of physical parameters. Moreover, we establish some properties of the solution(s). For the reader’s convenience, we provide in appendix a pedagogical and self-contained mathematical formulation of the DMFT equations for the finite Hubbard model. After recalling the definition and properties of one-body time-ordered Green’s functions and self-energies, and the mathematical structure of the Hubbard and Anderson impurity models, we describe the specific impurity solver we have studied in this article, namely the Iterated Perturbation Theory (IPT) solver, which can be conveniently formulated using Matsubara’s Green’s functions.

arXiv:2406.03384•Communications in Mathematical Physics•Jun 6, 2026•0 citations

Real-frequency diagrammatic Monte Carlo at finite temperature

J. Vučičević, Michel Ferrero

Diagrammatic expansions are a central tool for treating correlated electron systems. At thermal equilibrium, they are most naturally defined within the Matsubara formalism. However, extracting any dynamic response function from a Matsubara calculation ultimately requires the ill-defined analytical continuation from the imaginary- to the real-frequency domain. It was recently proposed [A. Taheridehkordi et al., Phys. Rev. B 99, 035120 (2019)] that the internal Matsubara summations of any interaction-expansion diagram can be performed analytically by using symbolic algebra algorithms. The result of the summations is then an analytical function of the complex frequency rather than Matsubara frequency. Here we apply this principle and develop a diagrammatic Monte Carlo technique which yields results directly on the real-frequency axis. We present results for the self-energy $\mathrm{\ensuremath{\Sigma}}(\ensuremath{\omega})$ of the doped $32\ifmmode\times\else\texttimes\fi{}32$ cyclic square-lattice Hubbard model in a nontrivial parameter regime, where signatures of the pseudogap appear close to the antinode. We discuss the behavior of the perturbation series on the real-frequency axis and in particular show that one must be very careful when using the maximum entropy method on truncated perturbation series. Our approach holds great promise for future application in cases when analytical continuation is difficult and moderate-order perturbation theory may be sufficient to converge the result.

arXiv:1908.11826•Physical review. B./Physical review. B•Feb 12, 2020•35 citations

Analytic Continuation of Electronic Green’s Functions from Imaginary to Real Time using Maximum Entropy

Johan Schött

This thesis is mainly a computational work studying the analytical continuation of Green’s functions using the Maximum Entropy method. A strongly correlated electron system is described with the single-band Hubbard model and paramagnetic solutions are studied using Dynamic Mean Field Theory on a Bethe lattice. Continuous Time Quantum Monte Carlo is used as Impurity solver, for the infinite Anderson model at a finite temperature, to obtain the Matsubara single-particle Green’s function propagator. Both metallic and insulating spectral functions are obtained using the Maximum Entropy Method. General properties of the Maximum Entropy Method as an analytic continuation method from imaginary to real time are also discussed.

•Jan 1, 2014•0 citations

Exact diagonalization as an impurity solver in dynamical mean field theory

Yi Lu, M. W. Haverkort

The dynamical mean-field theory (DMFT) maps a correlated lattice problem onto an impurity problem of a single correlated site coupled to an uncorrelated bath. Most implementations solve the DMFT equations using quantum Monte-Carlo sampling on the imaginary time and frequency (Matsubara) axis. We will here review alternative methods using exact diagonalization, i.e., representing the many-body ground state of the impurity as a sum over Slater determinants and calculating Green’s functions using iterative Lanczos procedures. The advantage being that these methods have no sign problem, can handle involved multi-orbital Hamiltonians (low crystal symmetry, spin-orbit coupling) and – when working completely on the real axis – do not need a mathematically ill-posed analytical continuation. The disadvantage of traditional implementations of exact diagonalization has been the exponential scaling of the calculation problem as a function of number of bath discretization points. In the last part we will review how recent advances in exact diagonalization can evade the exponential barrier thereby increasing the number of bath discretization points to reach the thermodynamic limit.

The European Physical Journal Special Topics•Jul 1, 2017•17 citations
We develop a novel analytic continuation method for self-energies on the Matsubara domain as computed by quantum Monte Carlo simulations within dynamical mean field theory (QMC+DMFT). Unlike a maximum entropy (maxEn) procedure employed for the last thirty years, our approach is based on a machine learning (ML) technique in combination with the iterative perturbative theory impurity solver of the dynamical mean field theory self-consistent process (IPT+DMFT). The input and output training datasets for ML are simultaneously obtained from IPT+DMFT calculations on Matsubara and real frequency domains, respectively. The QMC+DMFT self-energy on real frequencies is determined from the -- usually noisy -- input QMC+DMFT self-energy on the Matsubara domain and the trained ML kernel. Our approach is free from both, bias of ML training datasets and from fitting parameters present in the maxEn method. We demonstrate the efficiency of the method on the testbed frustrated Hubbard model on the square lattice.
A new algorithm for analytic continuation of noisy quantum Monte Carlo (QMC) data from the Matsubara domain to real frequencies is proposed. Unlike the widely used maximum-entropy (MaxEnt) procedure, our method is linear with respect to input data and can therefore be applied to off-diagonal components of a thermal Green's function, or to a self-energy function. The latter possibility is used to analyze QMC results for the half-filled single-band Hubbard model on a Bethe lattice at a low temperature. Our method qualitatively resolves peaks near the inner edges of the Hubbard bands in the vicinity of a Mott transition, whereas a MaxEnt procedure does not. An existence of such structures has been clearly established before in a high-precision D-DMRG calculation by Karski et al. We also analyze a stability of the new method subject to changes of adjustable parameters.
We show how to use diagrammatic techniques to compute the weak-coupling perturbation series of the self-consistent solution to a dynamical mean field theory (DMFT) problem. This approach constitutes an alternative to using diagrammatic techniques directly as an impurity solver. It allows one to bypass the need of multiple perturbative series resummations within the DMFT self-consistency loop. It can be applied at or out of equilibrium, with any diagrammatic formalism, such as real times, imaginary times, or Matsubara frequencies formalisms. As a proof of principle, we illustrate our method with the half-filled Hubbard model on the Bethe lattice in the DMFT approximation, using quantum quasi Monte Carlo (QQMC) to obtain the impurity perturbation series on the real time axis.
Abstract We provide a mathematical analysis of the Dynamical Mean-Field Theory (DMFT), a celebrated representative of a class of approximations in quantum mechanics known as embedding methods . Within the simplest setting of this approximation, which we introduce briefly at this point, we prove that, under certain assumptions, the DMFT equations admit a solution for any set of physical parameters. Moreover, we establish some properties of the solution(s). For the reader’s convenience, we provide in appendix a pedagogical and self-contained mathematical formulation of the DMFT equations for the finite Hubbard model. After recalling the definition and properties of one-body time-ordered Green’s functions and self-energies, and the mathematical structure of the Hubbard and Anderson impurity models, we describe the specific impurity solver we have studied in this article, namely the Iterated Perturbation Theory (IPT) solver, which can be conveniently formulated using Matsubara’s Green’s functions.
Diagrammatic expansions are a central tool for treating correlated electron systems. At thermal equilibrium, they are most naturally defined within the Matsubara formalism. However, extracting any dynamic response function from a Matsubara calculation ultimately requires the ill-defined analytical continuation from the imaginary- to the real-frequency domain. It was recently proposed [A. Taheridehkordi et al., Phys. Rev. B 99, 035120 (2019)] that the internal Matsubara summations of any interaction-expansion diagram can be performed analytically by using symbolic algebra algorithms. The result of the summations is then an analytical function of the complex frequency rather than Matsubara frequency. Here we apply this principle and develop a diagrammatic Monte Carlo technique which yields results directly on the real-frequency axis. We present results for the self-energy $\mathrm{\ensuremath{\Sigma}}(\ensuremath{\omega})$ of the doped $32\ifmmode\times\else\texttimes\fi{}32$ cyclic square-lattice Hubbard model in a nontrivial parameter regime, where signatures of the pseudogap appear close to the antinode. We discuss the behavior of the perturbation series on the real-frequency axis and in particular show that one must be very careful when using the maximum entropy method on truncated perturbation series. Our approach holds great promise for future application in cases when analytical continuation is difficult and moderate-order perturbation theory may be sufficient to converge the result.
This thesis is mainly a computational work studying the analytical continuation of Green’s functions using the Maximum Entropy method. A strongly correlated electron system is described with the single-band Hubbard model and paramagnetic solutions are studied using Dynamic Mean Field Theory on a Bethe lattice. Continuous Time Quantum Monte Carlo is used as Impurity solver, for the infinite Anderson model at a finite temperature, to obtain the Matsubara single-particle Green’s function propagator. Both metallic and insulating spectral functions are obtained using the Maximum Entropy Method. General properties of the Maximum Entropy Method as an analytic continuation method from imaginary to real time are also discussed.
The dynamical mean-field theory (DMFT) maps a correlated lattice problem onto an impurity problem of a single correlated site coupled to an uncorrelated bath. Most implementations solve the DMFT equations using quantum Monte-Carlo sampling on the imaginary time and frequency (Matsubara) axis. We will here review alternative methods using exact diagonalization, i.e., representing the many-body ground state of the impurity as a sum over Slater determinants and calculating Green’s functions using iterative Lanczos procedures. The advantage being that these methods have no sign problem, can handle involved multi-orbital Hamiltonians (low crystal symmetry, spin-orbit coupling) and – when working completely on the real axis – do not need a mathematically ill-posed analytical continuation. The disadvantage of traditional implementations of exact diagonalization has been the exponential scaling of the calculation problem as a function of number of bath discretization points. In the last part we will review how recent advances in exact diagonalization can evade the exponential barrier thereby increasing the number of bath discretization points to reach the thermodynamic limit.