Papers
Topics
Authors
Recent
Search
2000 character limit reached

Correctness criteria for complex Langevin

Published 14 Apr 2026 in hep-lat, cond-mat.str-el, and hep-th | (2604.12388v1)

Abstract: The complex Langevin approach is a promising method for the numerical treatment of systems with a sign problem, for which conventional lattice field theory techniques based on importance sampling cannot be applied. However, complex Langevin dynamics may fail to converge in some cases and converge to a wrong limit in others, motivating the development of various diagnostic tools over the years to assess the correctness of given simulation results. This work aims at providing a systematic comparison between the most prominent such correctness criteria. In particular, the main goal is to contrast their applicability, ease of use, and - most importantly - their predictive power. To this end, four simple but nontrivial models are considered and the criteria applied to each of them. The obtained conclusions are expected to carry over to more realistic theories as well.

Authors (1)

Summary

  • The paper compares several correctness diagnostics for complex Langevin simulations, focusing on convergence and analytic consistency.
  • It shows that combining analyticity and distributional criteria significantly improves the detection of sampling errors in sign-problematic models.
  • The study employs rigorous numerical tests across various models, paving the way for automated diagnostic frameworks in quantum field theory simulations.

Correctness Criteria for Complex Langevin: A Comparative Analysis

Introduction

The paper "Correctness criteria for complex Langevin" (2604.12388) investigates the diagnostic tools used to assess the validity of complex Langevin dynamics, a numerical technique for simulating systems afflicted by a sign problem where traditional importance sampling fails. The complex Langevin approach—while promising—has notable limitations, including non-convergence and convergence to incorrect limits. The core objective of the paper is to systematically compare the principal correctness criteria utilized for complex Langevin simulations, focusing on their applicability, usability, and predictive accuracy.

Complex Langevin Equation and Sign Problem

The complex Langevin equation (CLE) extends stochastic quantization to nonreal actions, enabling simulation of quantum field theories and statistical models with a sign problem, such as QCD at finite chemical potential. CLE replaces the real-valued drift term with its complex counterpart to circumvent the inability of importance sampling to deal with highly oscillatory weights. Nevertheless, CLE does not guarantee convergence to the correct expectation value, necessitating robust correctness criteria.

Review of Correctness Criteria

The work enumerates and scrutinizes several prominent correctness diagnostics:

  • The analyticity criterion: Probes whether observables remain analytic in parameters such as mass or coupling, exploiting holomorphicity to flag erroneous convergence.
  • The zero-mode criterion: Examines whether the average value of the drift vanishes in the stationary limit, which is required for correct sampling.
  • Distributional and boundary criteria: Inspects the behavior of probability distributions, especially at the boundaries of the complexified configuration space. Incorrect sampling is often reflected in pathological distributions or nonvanishing boundary terms.

Each criterion is assessed with respect to:

  • Its theoretical foundation based on the algebraic and analytic structure of CLE.
  • Practical tractability and ease of implementation in typical lattice field simulations.
  • Predictive power in identifying failure modes (non-convergence or convergence to the wrong result).

Simulation Setup and Model Selection

Four nontrivial models are chosen to rigorously evaluate the criteria: the one-dimensional quartic model, its variants, and other representative examples with varying degrees of sign problem severity and drift pathologies. These models allow for decisive exploration of the criteria’s strengths and weaknesses in controlled environments where exact results are available for comparison.

Comprehensive simulation pipelines are constructed using the Python scientific ecosystem and Wolfram Mathematica, allowing for high-precision numerical verification and reproducibility. Data sets and codebases are made publicly available for transparency and subsequent validation.

Comparative Evaluation and Numerical Results

Across all models, the paper finds:

  • Analyticity tests consistently identify incorrect convergence when CLE produces nonholomorphic expectation values, demonstrating significant predictive utility.
  • Boundary and distributional criteria are critical in spotting subtle drift-induced sampling errors, particularly those arising from insufficient control of excursions in the complexified configuration space.
  • Zero-mode criterion provides strong necessary but insufficient conditions; its violation confirms incorrectness, but its satisfaction does not guarantee correctness.

The paper quantitatively benchmarks each diagnostic, revealing that combining analyticity and distributional criteria offers superior reliability in flagging erroneous simulations. In several toy models, only these combined criteria are able to exclude improper sampling that escapes detection by other means.

Implications and Future Directions

The findings have immediate practical ramifications for numerical field theory: researchers must employ a hierarchy of correctness criteria, with analytic and distributional diagnostics forming the backbone of validation protocols. Theoretically, the results underscore the necessity of deeper analysis of the CLE’s boundary behavior and the complexification of configuration space, motivating further study of new correctness diagnostics and improved algorithms.

Future developments may include automated diagnostic frameworks leveraging the criteria evaluated here, applicable to more realistic quantum field theories and models with more intricate sign problems. Improved understanding of the relationship between the drift’s analytic structure and measure preservation could yield criteria that are both stricter and easier to implement.

Conclusion

Through systematic comparison of correctness criteria for complex Langevin dynamics, this paper demonstrates the strengths and limitations of analytic, distributional, and zero-mode diagnostics. Combining multiple criteria is shown to be indispensable for reliable assessment of simulation results. The work establishes practical guidelines for numerical studies of theories with sign problems and opens avenues for future development of CLE correctness diagnostics.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Collections

Sign up for free to add this paper to one or more collections.

Tweets

Sign up for free to view the 1 tweet with 2 likes about this paper.