⚛️ lattice

Complex Langevin kernels for real-time SU(N) lattice gauge theory in 3+1 dimensions

This paper introduces and validates novel kernel modifications for complex Langevin dynamics, enabling stable real-time simulations of 3+1 dimensional SU(2) lattice gauge theory up to unprecedented time scales while establishing a generalized correctness criterion to ensure convergence.

Original authors: Kirill Boguslavski, Paul Hotzy, David I. Müller

Published 2026-10-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Kirill Boguslavski, Paul Hotzy, David I. Müller

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the subatomic world, the forces that bind matter together are governed by rules that are notoriously difficult to calculate when time is involved. Physicists have long been able to map out the static properties of these forces, much like taking a photograph of a frozen moment. However, understanding how these forces evolve in real time—such as the behavior of the quark-gluon plasma created in high-energy particle collisions—requires a different kind of mathematics. The standard tools for these calculations rely on a mathematical trick that turns time into a spatial dimension, which works beautifully for static questions but fails completely when trying to track change. When researchers attempt to simulate these systems in real time, the equations produce a chaotic mix of positive and negative values that cancel each other out, making it impossible to extract a clear answer. This is known as the sign problem, and it has blocked a direct view into the real-time dynamics of the universe's most energetic events for decades.

To bypass this obstacle, a team of researchers has developed a new method to navigate these complex mathematical landscapes. They are using a technique called complex Langevin dynamics, which can be thought of as a way to walk through a foggy, multi-dimensional terrain to find the correct path. In this approach, the system is allowed to wander through a complex space where numbers have both real and imaginary parts. The challenge is that without careful guidance, this random walk tends to wander off into regions where the results are meaningless or unstable. The researchers' breakthrough lies in designing a set of steering mechanisms, or kernels, that gently guide the simulation back onto the correct path. By tailoring these guides to the specific geometry of the time-contour used in the simulation, they have managed to keep the process stable for much longer durations than ever before.

The team focused on simulating a theory of strong interactions known as SU(N) gauge theory, specifically looking at a version with two types of charges in a universe with three spatial dimensions and one time dimension. Their goal was to see how far they could push the simulation into real time before the numbers began to spiral out of control. Previously, using a standard guiding method, their simulations could only remain stable for a real-time duration equal to 1.5 times the inverse temperature of the system. In their new work, they introduced two distinct types of improvements to the steering mechanism. The first was a phase kernel, a guide that changes depending on where the simulation is along the time path but does not react to the specific state of the fields. This simple adjustment allowed them to extend the stable simulation time to three times the inverse temperature.

Pushing the boundaries further, the researchers then introduced a second, more sophisticated guide: a field-dependent kernel. Unlike the first version, this guide reacts to the local environment of the simulation, specifically looking at small loops of force fields called plaquettes to decide how to steer. This dynamic adjustment proved to be even more effective. By combining this local sensitivity with the geometric guidance of the time path, the team successfully stabilized the simulation for a duration of 3.5 times the inverse temperature. This represents a more than doubling of the accessible real-time window compared to their previous best results. To ensure that these longer simulations were not just stable but also correct, the researchers applied a rigorous test based on the behavior of the mathematical "drift" that drives the simulation. They verified that the probability of the system wandering into unstable regions dropped off exponentially, a mathematical signature that confirms the results are trustworthy.

The success of these new kernels was confirmed by comparing the simulation outputs against known benchmarks. When the researchers calculated specific properties of the force fields, such as the energy stored in loops of the field, the results from their long real-time simulations matched perfectly with results obtained from standard, stable methods that do not suffer from the sign problem. This agreement proves that the new steering mechanisms are not just preventing the simulation from crashing, but are actually guiding it to the physically correct answers. The researchers also noted that the optimal settings for these guides were not random; they found a clear relationship between the angle at which the time path was tilted in the complex plane and the specific phase settings required for the guide to work best.

This work marks a significant step forward in the ability to study non-Abelian gauge theories in real time. By extending the stable simulation window to 3.5 times the inverse temperature, the team has opened the door to calculating physical quantities that were previously out of reach, such as transport coefficients and spectral functions at low frequencies. These are the kinds of measurements needed to fully understand the behavior of the quark-gluon plasma and other extreme states of matter. The researchers emphasize that while their current results are based on simulations, the framework they have built provides a controlled and reliable way to explore these dynamics. Their findings suggest that with these new tools, it is now possible to perform first-principles calculations of real-time correlation functions at unprecedented time scales, bringing the theoretical understanding of the early universe and high-energy collisions closer to experimental reality. The next step for the team will be to apply these stabilized methods to compute unequal-time correlation functions, a task that will further test the limits of this new approach.

Technical Summary: Complex Langevin Kernels for Real-Time SU(N) Lattice Gauge Theory

Problem Statement
Real-time dynamics in quantum field theory, particularly for non-Abelian gauge theories like QCD, are essential for understanding phenomena such as the quark-gluon plasma. However, standard Lattice Gauge Theory (LGT) relies on Euclidean formulations, making direct access to real-time observables impossible due to the ill-posed nature of reconstructing Minkowski-space information from Euclidean correlators. Formulating lattice dynamics directly in Minkowski time leads to a highly oscillatory path-integral weight, resulting in a severe sign (or phase) problem that renders standard importance sampling ineffective.

The Complex Langevin (CL) method offers a potential solution by extending stochastic quantization to complex actions, allowing simulations on complex-time contours like the Schwinger-Keldysh (SK) contour. However, applying CL to non-Abelian SU(N) lattice gauge theories has historically been obstructed by "runaway trajectories" and convergence to incorrect results when the underlying dynamics are insufficiently controlled. While previous work introduced an anisotropic kernel that enabled stable simulations in 3+1 dimensions, the accessible real-time extent was limited (up to $t_{max} = 1.5\beta$ ), and stability degraded rapidly for longer times or larger couplings without drastic increases in lattice anisotropy.

Methodology
The authors investigate CL dynamics for real-time SU(N) gauge theories in 3+1 dimensions on a rectangular, regularized Schwinger-Keldysh time contour. The core methodology involves introducing and testing new kernel modifications tailored to lattice gauge theory to stabilize the CL process. The study employs a discretized CL equation for link fields $U_{x,\mu}$ on a hyper-cubic lattice, utilizing a higher-order Runge-Kutta-like stochastic solver to ensure second-order convergence, which is necessary for handling field-dependent diffusion terms.

The authors categorize their kernel approaches into two main classes:

Field-Independent Phase Kernels: These kernels depend on the position along the time contour but not on the gauge field configuration. They are constructed based on symmetry considerations (spatial homogeneity, locality, Euclidean neutrality, and time-reflection symmetry) to introduce complex phase factors that counteract the sign problem.
Field-Dependent Kernels: These kernels depend locally on the gauge field configuration, specifically through traced plaquette variables ( $Tr[U_{x,\mu\nu}]$ ). They are designed to be holomorphic, gauge-covariant, and local, modifying both the drift and diffusion terms of the Langevin equation to suppress excursions into non-unitary regions of the complexified configuration space.

To validate the correctness of the simulations, the authors generalize the correctness criterion proposed by Nagata et al. to kernelled dynamics. This criterion assesses the decay behavior of the probability density of the drift magnitude. Correct convergence requires this density to decay at least exponentially; power-law tails indicate a failure of the method.

Key Contributions

Generalization of Correctness Criteria: The paper extends the drift-magnitude-based correctness criterion to include kernelled CL evolutions, providing a practical a posteriori tool to validate convergence in the presence of kernel modifications.
Novel Kernel Constructions: The authors introduce the first application of specific field-independent phase kernels and local, field-dependent kernels (based on traced plaquettes) to 3+1 dimensional lattice gauge theories.
Higher-Order Solver Implementation: A higher-order stochastic solver compatible with field-dependent kernels is formulated and implemented, ensuring meaningful comparisons across different kernel settings.

Results
The authors demonstrate significant improvements in the stability and accessible real-time extent of CL simulations for pure SU(2) gauge theory at a bare coupling $g=0.5$ :

Baseline: Using the previously established anisotropic kernel on a rectangular SK contour, stable simulations are possible up to $t_{max} = 2\beta$ .
Phase Kernel: Introducing a contour-dependent phase kernel extends the stable real-time extent to $t_{max} = 3\beta$ . The drift magnitude distributions for both temporal and spatial components exhibit exponential decay, satisfying the correctness criterion.
Field-Dependent Kernel: A local kernel based on traced plaquettes further extends the reach to $t_{max} = 3.5\beta$ . This configuration also yields exponential decay of the drift magnitude and maintains a bounded unitarity norm over long Langevin times.
Validation: The results obtained with the new kernels are cross-validated against Euclidean benchmarks. Planar spatial Wilson loops computed on the real-time contour agree with Euclidean results within statistical uncertainties, confirming that the simulations yield physically correct expectation values.

Significance and Claims
The paper claims to represent the first application of such kernel modifications to lattice gauge theories in 3+1 dimensions. The results substantially advance the applicability of the Complex Langevin method to real-time simulations in Minkowski space, establishing a framework for controlled studies of non-Abelian gauge theories at unprecedented time scales (more than doubling the previous limit of $1.5\beta$ ).

The authors emphasize that these developments provide a pathway to compute unequal-time correlation functions and spectral information at low frequencies, which are currently inaccessible via standard methods. They note that while the current work focuses on pure gauge theory, the framework is designed to be applicable to more complex observables. The paper concludes that the strategic use of kernels, guided by symmetry principles, offers a practical and adaptable mechanism to diagnose and correct pathological behavior in CL simulations, bridging a critical gap between lattice gauge theory and real-time physics. A forthcoming paper is cited as applying these developments to the computation of unequal-time correlation functions.

Technical Summary: Complex Langevin Kernels for Real-Time SU(N) Lattice Gauge Theory

More like this