Origin of misleading convergence in self-consistent many-electron theories: Fundamental aspects and practical implications

Herbert Eßl, Matthias Reitner, Evgeny Kozik, Alessandro Toschi · 2025

Read Origin of misleading convergence in self-consistent many-electron theories: Fundamental aspects and practical implications by Herbert Eßl, Matthias Reitner, Evgeny Kozik, Alessandro Toschi (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-2502-01420.txt.

Introduction

Essl and co-authors probe in 2025 why some physics solvers settle on a false answer for systems of many electrons. Such self checking methods are a main tool for strongly linked materials. In those hard cases, the solver often converges, but to a state that cannot be real.

The team derives the math rule for when the true state stays stable in the solver. That rule links false settling to a split valued core function, though false settling can occur without the linked warning signs. They end with a step by step way to hold the solver on the true state. The work has clear use for groups that model strong electron systems.

Catalog record

CREATORHerbert Eßl, Matthias Reitner, Evgeny Kozik, Alessandro Toschi
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-01420

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Origin of misleading convergence in self-consistent many-electron theories: Fundamental aspects and practical implications Herbert Eßl, Matthias Reitner, Evgeny Kozik, Alessandro Toschi Abstract Self-consistent approaches in many-electron problems typically converge to an unphysical solution in strongly correlated regimes. By deriving the mathematical condition for the stability of the physical solution, we unveil the precise relation between two distinct issues previously considered equivalent: the misleading convergence in self-consistent schemes and the multivaluedness of the Luttinger-Ward functional. Although these problems are fundamentally linked through the divergences of the irreducible vertex function, we show that misleading convergence can occur even in the absence of such divergences. Eventually, a systematic procedure for stabilizing the physical solution is proposed. Introduction – Many fascinating phenomena in condensed matter physics, such as unusual spectral features (pseudogap [ 1 , 2 ] , waterfalls [ 3 , 4 ] , kinks [ 5 ] , etc.), unconventional superconductivity [ 6 , 7 , 8 , 9 ] , quantum criticality [ 10 , 11 , 12 , 13 ] , and Mott/Hund’s metal-insulator transitions [ 14 , 15 , 16 , 17 ] are driven by electronic correlations. This sets, for their theoretical description, the high bar of solving a full quantum many-body problem of interacting fermions. While a Feynman diagrammatic formalism could be employed, this lacks, unlike in QED, a small expansion parameter. Thus, for describing the interesting physics mentioned above, Feynman diagrammatic calculations often need to be performed in the non-perturbative regime. This problem affects the important class of conserving Baym-Kadanoff approximations, constructed through self-consistent resummations of Feynman diagrams, which guarantee the thermodynamic “consistency” of the results [ 18 , 19 , 20 , 21 , 22 , 23 ] . Formally, self-consistent diagrammatic methods compute the electron Green’s function G as a functional f of G itself, with the implicit equation G=f[G] being solved by iterations. With the exception of extremely simple models the exact map f is unknown, but can be constructed approximately, e.g. as truncated diagrammatic series in terms of G , summed to high order by the diagrammatic Monte Carlo technique [ 24 , 25 ] , the GW scheme [ 26 , 27 ] , or, non-perturbatively, using dynamical mean-field theory [ 28 ] and its extensions [ 29 , 30 ] , nested-cluster [ 31 ] or parquet-based [ 32 ] approaches. In the last decade, an increasing number of studies have reported that an iterative solution of G=f[G] can converge to a wrong answer [ 33 , 31 , 34 , 35 , 5 , 37 , 38 ] , typically in the most interesting correlated regimes. These studies have ascribed the occurrence of misleading convergence in the iterative solution for G to the intrinsic multivaluedness of the Luttinger-Ward functional (LWF) of the many-electron problem [ 1 , 2 , 34 , 23 ] . Any crossing of two branches of the LWF (with only one being physical) 1 1 1 Precisely, it was proven that the crossing of two solutions of the equation G[G_{0}]=G_{\text{phys}} implies a vertex divergence. However, the existence of multiple solutions for this equation originates from the multivaluedness of the LWF. implies the simultaneous divergence of the two-particle irreducible (2PI) vertex function [ 42 , 43 , 5 , 44 , 31 , 45 , 7 , 47 , 48 , 4 ] . Physically, this divergence is intrinsically linked [ 44 , 50 , 51 , 52 ] to one of the most relevant feats of electronic correlation: the local magnetic moment formation and the associated freezing of charge fluctuations, as this huge differentiation of on-site magnetic and charge responses requires extremely large vertex functions. Notably, vertex divergences also yield the necessary condition [ 53 , 54 , 55 , 56 , 57 ] for the onset of phase-separations in the proximity of Mott-Hubbard metal-insulator transitions [ 14 , 58 , 59 , 60 ] . More formally, vertex divergences represent a key hallmark [ 6 , 53 , 50 , 62 , 51 , 63 , 52 ] of the breakdown of self-consistent perturbation theory (SCPT). Specifically, the skeleton diagrammatic series in terms of G was found [ 34 , 64 , 65 ] to always converge to an unphysical answer for interaction values larger than those at which the first branching point of the LWF and, hence, the first vertex divergence, occurs. However, so far, the evidence for the connection between the misleading convergence of the iterative solution of G=f[G] and the crossings between the LWF branches has remained empirical . This is likely the reason, why the misleading convergence issues affecting GW [ 35 ] , parquet-based [ 37 ] , and nested-cluster schemes [ 31 ] , often in the proximity of vertex divergences, have been hitherto attributed to the branching of the LWF, even if these iterative approaches, differently from SCPT, are not based on perturbative series of LWF diagrams. At the same time, misleading convergences have also been reported in regions of parameter space without such crossings, i.e. vertex divergencies [ 34 , 31 , 65 ] . Evidently, a complete understanding of this problem and its implications is still lacking. In this work, we aim to clarify the origin of the misleading convergence observed in iterative many-body schemes, formally disentangling this issue from the branching of the LWF. To this end, we first devise a rigorous, model-independent procedure to determine, whether the fixed points of the map f are stable or unstable. This allows us to prove that, for the same parameters at which branch crossings of the LWF (and the associated vertex divergences) occur, the physical fixed point of the iterative procedure also becomes unstable. However, as we will also demonstrate, in absence of vertex divergences, both the misleading convergence and the breakdown of the skeleton series can still occur but no longer simultaneously . Through our analytical derivation, we will be able to eventually unveil the precise relation between these two important issues of the self-consistent theories for the many-electron problem. Finally, we derive a systematic approach for changing the map f (to a modified map \tilde{f} ) that stabilizes the physical fixed point (see the illustration in Fig. 1 ), thereby ensuring convergence to the correct physical solution over the entire parameter range. Models – Our analytic theory will be illustrated by its application to the zero point (ZP) model, for which the functional f simplifies to a function, and to the Hubbard atom (HA). These are defined by the action \displaystyle S=-\sum_{\nu,\sigma}\bar{c}_{\nu,\sigma}G_{0,\nu}^{-1}c_{\nu,\sigma}+S_{\text{int}}, (1) where \nu is a fermionic Matsubara frequency, \bar{c}(c) are Grassmann variables that represent the creation (annihilation) of an electron (see appendix A) and G_{0,\nu}^{-1}=i\nu+\delta\mu with \delta\mu being the chemical potential ( \delta\mu=0 corresponds to ph-symmetry). The interacting part of the action reads S_{\text{int,ZP}}=\frac{U}{\beta}\sum_{\nu}\bar{c}_{\nu,\uparrow}c_{\nu,\uparrow}\bar{c}_{\nu,\downarrow}c_{\nu,\downarrow} for the ZP model and S_{\text{int,HA}}=U\int_{0}^{\beta}d\tau\,(\bar{c}_{\tau,\uparrow}c_{\tau,\uparrow}-1/2)(\bar{c}_{\tau,\downarrow}c_{\tau,\downarrow}-1/2) for the HA, with U being the on-site interaction, \beta the inverse temperature and \tau the imaginary time. These basic models are selected for illustrative purposes and for a rigorous benchmark w.r.t. their exact solutions. Yet, they still entail, in a crude fashion, pivotal feats of the electron-correlation physics. The HA, which describes totally localized electrons, yields a somewhat “extreme representation” of

Read the complete text