Real-Time Functional Renormalization Group

• Real-time FRG is a method that directly computes dynamical observables (e.g., spectral functions, decay widths) without relying solely on Euclidean correlation functions.
• It employs both analytic continuation and real-time contour formulations to access nonequilibrium dynamics and resolve causal response features.
• Regulator choices and truncation schemes are pivotal for accurately preserving symmetries and extracting physical insights in applications such as hadronic and critical dynamics.

Searching arXiv for recent and foundational papers on real-time functional renormalization group. Real-time functional renormalization group (real-time FRG) denotes a family of functional-renormalization-group constructions designed to compute dynamical observables—retarded propagators, spectral functions, decay widths, transport coefficients, pole masses, and dynamic critical exponents—without reducing the problem to a purely Euclidean one and postponing Minkowski information to a numerically ill-posed reconstruction step. In the literature, this goal is pursued mainly in two ways: by analytically continuing FRG flow equations themselves from Matsubara frequencies to real frequencies, and by formulating the flow directly on real-time contours such as Schwinger–Keldysh or, for stochastic dynamics, Martin–Siggia–Rose–Janssen–de Dominicis (MSRJD) (Kamikado et al., 2013, 0809.5208, Huelsmann et al., 2020).

1. Conceptual scope

Standard FRG is usually organized around the Wetterich equation for a scale-dependent effective average action, but in its customary Euclidean form it naturally delivers static quantities and Euclidean correlation functions. Real-time FRG addresses the mismatch between that formulation and the fact that physical thresholds, branch cuts, widths, and causal response are encoded in retarded correlators $G_R(\omega,\vec p)$, not in Euclidean correlators at Matsubara frequencies. Taken together, the literature indicates that “real-time FRG” is best understood as a methodological umbrella rather than a single formalism.

Strand	Core construction	Typical targets
Analytically continued flow equations	Continue external frequency in the FRG flow before solving	Spectral functions, decay widths, pole masses
Closed-time-path or MSRJD FRG	Double fields on a real-time contour or stochastic response contour	Nonequilibrium fixed points, critical dynamics, transport
Periodic steady-state extensions	Reformulate nonequilibrium Keldysh-fRG in Floquet space	Long-time driven quantum-dot observables

This suggests a basic division of labor. Analytically continued flows are tailored to equilibrium or near-equilibrium linear response, where Matsubara analyticity can still be exploited. Contour-based flows are needed once the statistical and spectral sectors become independent dynamical objects, as in genuine nonequilibrium evolution, dynamic critical phenomena with reversible mode couplings, or frequency-dependent transport beyond equilibrium (Pawlowski et al., 2015, Berges et al., 2012, Eissing et al., 2016).

2. Analytic continuation at the level of FRG flows

A central line of development replaces post hoc continuation of Euclidean data by continuation of the flow equations themselves. In the scalar $O(N)$ model, a practically usable scheme was constructed by deriving the two-point-function flows, approximating higher vertices by momentum-independent derivatives of the scale-dependent effective action, and continuing the external Euclidean frequency according to the retarded prescription

$$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$

With a three-dimensional regulator, the continuation is straightforward because no extra regulator poles appear in the complex $p_0$-plane at finite $k$; the resulting truncation is tied consistently to the derivative expansion and reproduces the potential curvatures at $p=0$, so that the pion remains a Nambu–Goldstone mode in the chiral limit (Kamikado et al., 2013).

A second step generalized this strategy to Lorentz-invariant regulators and Minkowski-space derivative expansions. In this approach, the inverse propagator is parameterized near the real axis as

$P(p)=P_1(p^2)-i\,s(p_0)\,P_2(p^2),$

with $s(p_0)=\operatorname{sign}(\Re p_0\,\Im p_0)$, and the flow is projected directly onto on-shell quantities such as residues and width parameters. In the broken $O(N)$ model this yields flow equations for the radial-mode discontinuity and hence for the decay width into Goldstones, while preserving Lorentz or Galilei invariance at the truncation level through suitable regulator choices (Floerchinger, 2011, Riebesell, 2017).

A further generalization computes correlation functions for complex external frequencies and reconstructs retarded correlators by explicit pole corrections. Rather than assuming that the analytically continued Euclidean loop expression already equals the retarded one, this framework tracks the residue contributions that distinguish them. Its regulator construction is explicitly designed to preserve spacetime symmetries while keeping regulator poles outside the strip of complex frequencies relevant for the target real-time observables (Pawlowski et al., 2015).

The same continuation philosophy was then extended across models and statistics. In-medium sigma and pion spectral functions in the quark-meson model, including finite temperature, quark chemical potential, and nonzero spatial momentum, were obtained from analytically continued two-point flows; the same proceedings also applied the method to vector and axial-vector channels in a gauged linear sigma model (Wambach et al., 2017). Fermionic continuation was implemented for the vacuum quark propagator in the quark-meson model through a Dirac decomposition of the retarded two-point function, yielding quark spectral functions and explicit threshold structure without numerical reconstruction (Tripolt et al., 2018). In a nonrelativistic setting, the mobile impurity in a Fermi sea was treated by solving a Euclidean derivative expansion for running couplings and then integrating analytically continued flow equations for the full retarded inverse propagators, producing attractive and repulsive polarons together with a molecule-hole continuum (Kamikado et al., 2016).

3. Real-time contour formulations

A more direct route starts from a real-time contour from the outset. On the Schwinger–Keldysh closed time path, the fields are doubled and the statistical correlator $F$ and spectral function $O(N)$0 are independent objects. This is the decisive structural difference from equilibrium Euclidean FRG, where fluctuation-dissipation or KMS relations tie the two sectors together. In the Keldysh or classical/quantum basis, the exact real-time flow equation becomes a one-loop exact functional equation involving full retarded, advanced, and statistical propagators rather than a single Euclidean propagator (0809.5208, Berges et al., 2012).

Within that framework, nonthermal fixed points in relativistic $O(N)$1 theory were formulated as genuine real-time RG fixed points. The fixed-point condition can be written as the stationarity identity

$O(N)$2

which is the real-time gain-minus-loss balance familiar from kinetic theory. In a resummed large-$O(N)$3 truncation this leads to the nonequilibrium scaling branches

$O(N)$4

thereby placing vacuum, thermal, and nonthermal fixed points within one common RG hierarchy (0809.5208, Berges et al., 2012).

A detailed Schwinger–Keldysh FRG construction for scalar spectral functions was later developed in a form suitable for numerical solution. There the central lesson is truncational: in the symmetric phase, a local-potential approximation cannot generate a finite width, because the relevant imaginary part of the retarded self-energy first appears through nonlocal sunset-type structures. The practical implication is that real-time spectral broadening requires nonlocal four-point vertices in the truncation; with such vertices included, the propagator flow becomes two-loop complete and reproduces benchmark spectral functions in $O(N)$5 dimensions rather well (Huelsmann et al., 2020).

For stochastic critical dynamics with reversible mode couplings, the real-time contour language is replaced by an MSRJD formulation. In the dynamic universality class relevant to the two-flavor chiral transition in the chiral limit, the slow fields are an $O(N)$6 order parameter and conserved antisymmetric charge densities coupled through non-Abelian Poisson brackets. In that setting the regulator is introduced through the free energy and couples to composite response fields, a construction used to preserve causality, detailed balance, conservation laws, and a temporal gauge or displacement symmetry that implies non-renormalization of the reversible mode coupling. The resulting real-time FRG flow reproduces the strong-scaling prediction

$O(N)$7

and yields a universal scaling function for the diffusion coefficient (Roth et al., 2024).

4. Truncations, regulators, and symmetry constraints

All real-time FRG implementations confront the same structural issue: differentiating the Wetterich equation generates an infinite hierarchy in which the flow of an $O(N)$8-point function depends on higher vertices. The practical literature therefore uses closures matched to specific approximation schemes. In the analytically continued $O(N)$9 calculation, higher vertices in the two-point flow are replaced by field derivatives of the scale-dependent potential, which keeps the truncation consistent with the local potential approximation and ensures that, at $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$0, the two-point functions reproduce the potential curvatures exactly (Kamikado et al., 2013). In the quark-meson and gauged linear sigma applications, the same LPA logic is used, with Yukawa vertices fixed by a constant coupling and higher derivative operators neglected (Wambach et al., 2017). The Fermi-polaron calculation instead imports scale-dependent parameters from a Euclidean derivative expansion into the full real-frequency two-point flow as a one-step improvement beyond the derivative expansion (Kamikado et al., 2016).

Regulator choice is equally decisive. Three-dimensional regulators simplify analytic continuation because they leave the internal frequency dependence simple enough for explicit Matsubara summation and avoid extra poles in the complex external-frequency plane; this is why they recur in the early mesonic, fermionic, and in-medium spectral-function papers (Kamikado et al., 2013, Wambach et al., 2017, Tripolt et al., 2018). The price is explicit breaking of Euclidean $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$1 or Minkowski Lorentz invariance at finite cutoff. By contrast, symmetry-preserving four-dimensional regulators retain Lorentz structure but introduce regulator poles that must be shifted out of the relevant strip of complex frequencies, for example by an auxiliary mass deformation $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$2 (Pawlowski et al., 2015). In the Schwinger–Keldysh spectral-function program, a causal $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$3-dimensional regulator is constructed so that retarded analyticity and equilibrium symmetry remain explicit (Huelsmann et al., 2020). In the MSRJD critical-dynamics framework, the regulator is introduced through the free energy and thereby preserves detailed balance and the static sector’s decoupled Euclidean flow (Roth et al., 2024).

A recurring technical theme is that real-time truncations must respect symmetry constraints more stringently than static Euclidean ones. Goldstone protection at $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$4, fluctuation-dissipation relations, Ward identities associated with conserved charges, and causal analyticity conditions all appear in the supplied literature as nontrivial consistency tests rather than optional refinements.

5. Representative applications

The most developed application class is hadronic and chiral dynamics. In the broken-phase $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$5 model, the direct real-time flow resolves the $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$6 threshold, generates a broad sigma enhancement above $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$7, and shows that the infrared sigma feature is predominantly dynamical in origin rather than a simple downward renormalization of the ultraviolet sigma pole (Kamikado et al., 2013). In the quark-meson model and its vector-extended variant, analytically continued flows at finite $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$8, $$\Gamma^{(2)R}_{k,j}(\omega)\equiv \Gamma^{(2)R}_{k,j}\bigl(p_0=-i(\omega+i\epsilon),\vec p=0\bigr).$$9, and $p_0$0 produce in-medium spectral functions for $p_0$1 and $p_0$2, including chiral-partner degeneracy at high temperature and the broadening-$p_0$3, lowering-$p_0$4 pattern discussed near the crossover (Wambach et al., 2017). The same method, applied to the vacuum quark propagator, yields fermionic spectral functions with explicit quark-pion and quark-sigma thresholds, a stable quasiparticle pole below threshold, and a clear distinction between pole and curvature masses (Tripolt et al., 2018).

A second major domain is nonequilibrium and critical dynamics. The real-time FRG on the Schwinger–Keldysh contour provides a common language for vacuum, thermal, and nonthermal fixed points in the relativistic $p_0$5 model, including the turbulence-related exponents $p_0$6 and $p_0$7 (0809.5208). In thermal equilibrium, a real-time FRG for the relativistic $p_0$8 model was used to follow the crossover from coherent relativistic dynamics with $p_0$9 at $k$0 to dissipative relaxational dynamics at finite temperature, with the infrared flow reducing, within the truncation, to Model A and $k$1 (Mesterházy et al., 2015). For the dynamic universality class of the two-flavor chiral transition with reversible mode couplings, the MSRJD-based real-time FRG recovers the strong-scaling exponent $k$2 and extracts a universal scaling function for charge diffusion (Roth et al., 2024).

Condensed-matter and mesoscopic applications show that the methodology is not confined to relativistic bosons. In the Fermi-polaron problem, the real-frequency flow resolves attractive and repulsive polaron peaks and a molecule-hole continuum, features inaccessible to a simple derivative expansion (Kamikado et al., 2016). In the nonequilibrium Kondo problem, a Keldysh-contour FRG was built for finite-frequency current noise; its distinctive ingredient is a time-nonlocal current vertex $k$3, required for a current-conserving treatment and responsible for anti-resonances in the noise spectrum at $k$4 and corresponding peaks in the ac conductance (Moca et al., 2010). For periodically driven interacting quantum dots, Keldysh-fRG was reformulated in Floquet space to access the long-time periodic regime, where the driving frequency $k$5 acts as an infrared cutoff and generates new power laws in renormalized hopping harmonics and pumped currents (Eissing et al., 2016). At the simplest end of the spectrum, a real-time FRG study of the quantized $k$6 anharmonic oscillator in $k$7 dimensions found only a single symmetric phase, a negligible field-independent wavefunction renormalization, and a noticeable field-dependent one, while also demonstrating substantial scheme dependence at fixed perturbative truncation order (Nagy et al., 2010).

6. Limitations, distinctions, and recurrent misconceptions

The supplied literature is unusually explicit about limitations. Analytic-continuation schemes remain primarily equilibrium or near-equilibrium linear-response methods; their strength is to avoid numerical reconstruction from Euclidean data, but they do not replace a genuine nonequilibrium contour formalism when fluctuation-dissipation relations are absent (Pawlowski et al., 2015, Floerchinger, 2011). Simple truncations often omit running wavefunction renormalizations, momentum dependence of higher vertices, back-coupling of momentum-dependent propagators into the effective potential, self-consistent vector loops, or six-point functions; these omissions show up as pole-curvature mass mismatches, missing widths, regulator dependence, or spurious spectral features (Wambach et al., 2017, Huelsmann et al., 2020, Kamikado et al., 2016).

Regulators are a second persistent source of tension. Three-dimensional regulators simplify continuation but break Lorentz or $k$8 symmetry at finite $k$9; four-dimensional regulators preserve symmetry more faithfully but introduce regulator poles that must be controlled. In stochastic real-time formulations, even the source and regulator sectors must be chosen with care, because coupling sources directly to elementary response fields can violate detailed balance when reversible mode couplings are present (Pawlowski et al., 2015, Roth et al., 2024).

A recurrent misconception concerns terminology. The “real-space RG” of pseudofermion FRG for SU($p=0$0) Heisenberg magnets is unrelated to real-time FRG; there, “real-space” refers to lattice-site dependence of vertices, not to Minkowski time or nonequilibrium dynamics (1711.02182). Another distinction is methodological rather than semantic: Floquet-space fRG treats asymptotic periodic steady states, not the full transient time evolution after switch-on (Eissing et al., 2016). Taken together, the literature suggests that real-time FRG is best viewed as a family of closely related strategies whose common objective is direct access to dynamical correlation functions while preserving analyticity, causality, and symmetry constraints as far as the truncation allows (Berges et al., 2012).