In quantum many-body physics and quantum field theory, there is a persistent tension between mathematical stability and physical reality.
When calculating the properties of strongly interacting systems—such as the quark-gluon plasma or strongly correlated electronic materials—the most robust numerical tools operate in Euclidean space (imaginary time, $\tau$, or discrete Matsubara frequencies, $i\omega_n$). In Euclidean space, quantum path integrals behave like classical statistical mechanics problems, free from the highly oscillatory phase factors that cause numerical "sign problems."
However, we do not live in imaginary time. The physical observables we measure in experiments—such as spectral functions, transport coefficients, scattering amplitudes, and sharp quasiparticle dispersion kinks—exist in Minkowski space (real time, $t$, or real frequency, $\omega$).
To bridge this gap, physicists traditionally rely on analytic continuation: computing a correlation function numerically in Euclidean space, and then mapping it to the real axis via the mathematical substitution $i\omega_n \to \omega + i\eta$.
Traditional Approach (Post-Hoc Continuation):
[Microscopic Theory]
│
▼ (Stable Numerical Integration)
[Euclidean Data G(iω)] ───► [Numerical Inversion (Ill-Posed)] ───► [Real-Frequency Spectrum ρ(ω)]
(Blurred/Unstable)
This post-hoc transformation is a notoriously ill-posed inverse problem. Because a continuous real-frequency spectrum is inferred from a finite set of noisy imaginary-time data points, tiny numerical uncertainties explode into massive errors on the real axis. As noted by Samuel Lederer et al. and Juan Yao et al., this inversion acts as a numerical blur, washing out sharp, vital dynamical features.
To escape this trap, a powerful paradigm has emerged: analytically continuing the renormalization group flow equations themselves, rather than continuing the final numerical data. By embedding the real-frequency transformation directly into the differential equations that define the theory, researchers can compute real-time spectral functions directly, bypassing the ill-posed inverse problem entirely.
Modern Approach (Flow Equation Continuation):
[Microscopic Theory]
│
▼ (Analytic Continuation of the Flow Equations)
[Real-Frequency Differential Equations]
│
▼ (Direct Integration)
[Real-Frequency Spectrum ρ(ω)] (Sharp & Precise)
The Core Mechanism: Shifting Continuation to the Differential Level
The primary vehicle for this approach is the Functional Renormalization Group (FRG). The FRG framework tracks how a physical system's effective action evolves across different energy scales, moving from an un-correlated ultraviolet (UV) scale down to the macroscopic infrared (IR) scale via differential flow equations.
In traditional FRG applications, these differential equations are integrated along the imaginary frequency axis. The breakthrough realized by Ralf-Arno Tripolt et al. and Kazuhiko Kamikado et al. was to perform the analytic continuation inside the loop integrals of the flow equations before running the numerical simulation.
Instead of tracking discrete Matsubara frequencies, the flow equations for the two-point correlation functions are formulated directly with a continuous real energy variable ($\omega + i\eta$). As the renormalization scale $k$ steps downward, the differential equations calculate exactly how the real-frequency spectral features grow, split, and interact.
Why does this bypass the ill-posedness? In the traditional approach, you solve an integral equation of the form:
$$G(\tau) = \int K(\tau, \omega) \rho(\omega) d\omega$$
where the kernel $K(\tau, \omega)$ exponentially suppresses high-frequency information, making inversion unstable. In the continued flow equation approach, you are instead solving an initial value problem for a set of coupled differential equations. The analytic structure of the propagator is propagated smoothly and deterministically throughout the flow, allowing sharp physical features to emerge naturally without numerical amplification of noise.
Self-Consistency in the Complex Momentum Plane
While substituting $i\omega_n \to \omega + i\eta$ into the flow equations sounds straightforward, it introduces a major complication: as the renormalization scale $k$ changes, the particles' self-energies change, causing the poles and branch cuts of the propagators to move across the complex plane. If a pole crosses the contour of integration during the flow, the calculation breaks down.
To resolve this, Nils Strodthoff developed a method to solve the FRG flow equations directly on a grid spanning the complex momentum and frequency plane.
Instead of avoiding the poles, the algorithm explicitly tracks their locations and maps the evolution of propagators and vertices across the complex grid. Crucially, the author introduced a back-coupling mechanism: real-time pole corrections are fed back into the flow equations at every step. This ensures that the calculated spectral functions are entirely self-consistent, capturing the exact feedback loop between a particle's real-time lifetime (imaginary part of the pole) and its effective mass (real part of the pole).
Convergence with Real-Time Formalisms
The philosophy of bypassing post-hoc analytic continuation by working directly on the real axis is not unique to the FRG; it represents a broader convergence across quantum many-body physics.
For instance, the Schwinger-Keldysh closed-time-path (CTP) formalism—long used in non-equilibrium field theory by Fred Cooper et al. and Nicolai Christiansen et al.—uses a forward and backward temporal loop to derive strictly causal, real-valued equations of motion directly in real time.
A key theoretical insight by Jan M. Pawlowski et al. demonstrated that the complex-frequency continuation of Euclidean flows is formally equivalent to this real-time Keldysh formalism. By using spatial momentum regulators that preserve causality, one can seamlessly map regulated complex-frequency flows to real-time integration paths.
This same drive to operate directly on the real axis has inspired alternative non-perturbative methods:
- Distributional Exact Diagonalization (Dist-ED): Developed by Mats Granath et al., this method solves quantum impurity models directly on the real-frequency axis by stochastically averaging over small, exactly solvable configurations, preserving sharp dispersion kinks that traditional methods erase.
- Real-Time Inchworm QMC: As implemented by Qiaoyuan Dong et al., this method uses a complex time path to simulate the real-time evolution of dynamical mean-field equations, dodging the dynamical sign problem to observe quasiparticle peak formations directly.
Summary
By shifting analytic continuation from a post-processing step to a foundational property of the differential flow equations, researchers have unlocked a way to calculate real-time dynamics from first principles. Whether tracking moving poles across a complex momentum grid or integrating along a Keldysh contour, the core intuition remains the same: to understand the real-time dynamics of strongly correlated quantum systems, we must let the physics flow along the real axis from the very start.
Key Foundational Work
Nonequilibrium quantum fields in the large-Nexpansion (1994) — contributed 3 concept elements to this direction.
Distributional exact diagonalization formalism for quantum impurity models (2012) — contributed 2 concept elements to this direction.