---
title: Nonequilibrium Bosonic DMFT
url: https://www.emergentmind.com/topics/nonequilibrium-bosonic-dynamical-mean-field-theory
type: topic
---

# Nonequilibrium Bosonic DMFT

Searching arXiv for recent and foundational papers on nonequilibrium bosonic dynamical mean-field theory, including open-system extensions.
Nonequilibrium bosonic dynamical mean-field theory (BDMFT) is a nonequilibrium extension of bosonic dynamical mean-field theory for strongly interacting lattice bosons, formulated so that local quantum dynamics is treated nonperturbatively while lattice feedback enters through a self-consistent effective field and, in the full DMFT formulation, a dynamical hybridization function [1405.6941, 2008.02563]. In the Bose–Hubbard setting, the construction yields an effective single-site impurity problem on a real-time contour with full Nambu structure, allowing one to address normal and Bose-condensed phases, damping, relaxation, and thermalization beyond Gutzwiller mean-field theory [1405.6941]. In Markovian open quantum systems, the same logic maps a Lindblad lattice problem onto a dissipative impurity embedded in a self-consistent coherent field and a non-Markovian bath, thereby incorporating finite-connectivity corrections beyond Gutzwiller and revealing hopping-induced dissipative processes, finite-frequency instabilities, and stationary quantum-Zeno behavior [2008.02563]. A distinct large-dimensional perspective rigorously justifies the simplest Weiss-field closure, showing that in the high-dimensional limit the one-site reduced dynamics converges to a nonlinear, strongly interacting single-site evolution in which the mean field acts in the hopping channel and not in the interaction [2501.05304].

## 1. Conceptual setting and scope

Nonequilibrium BDMFT is designed for bosonic lattice systems whose dynamics is inaccessible to static mean-field treatments because the relevant physics depends on memory, retardation, and the feedback of local fluctuations onto the lattice environment [1405.6941]. The formalism is especially natural for the Bose–Hubbard model, where the local interaction remains fully nonperturbative while the intersite hopping is encoded through a self-consistent bath or Weiss field [1405.6941, 2501.05304].

The nonequilibrium formulation in closed systems is built on the Kadanoff-Baym contour
\[
\mathcal{C}=(0\rightarrow t_{\max}\rightarrow 0\rightarrow -i\beta),
\]
which permits finite-temperature initial conditions and real-time evolution within a single contour-ordered framework [1405.6941]. In open systems, the formalism is recast in Lindbladian Keldysh language, because dissipation is local and Markovian at the microscopic level while non-Markovian memory emerges self-consistently through the DMFT bath [2008.02563].

A central organizing distinction is between two levels of theory. The full nonequilibrium BDMFT construction uses a dynamical hybridization function \(\mathbf{\Delta}(t,t')\), typically related self-consistently to the local Green’s function, and thereby captures damping and thermalization beyond site-decoupled dynamics [1405.6941, 2008.02563]. By contrast, the high-dimensional rigorous result establishes the validity of the simpler Weiss-field equation
\[
i \partial_t \varphi(t) = h^{\varphi(t)} \varphi(t),
\]
with a self-consistent order parameter \(\alpha_\varphi=\langle \varphi,a\varphi\rangle\), but without a dynamical hybridization bath [2501.05304]. This implies that “nonequilibrium bosonic DMFT” may refer either to the full Keldysh impurity theory with hybridization or to its large-dimensional Weiss-field limit, depending on context.

## 2. Lattice models and nonequilibrium formulations

For closed bosonic lattices, the canonical model is the Bose–Hubbard Hamiltonian
\[
H = -J \sum_{\langle i,j \rangle} \big(b_i^\dagger b_j + b_j^\dagger b_i\big)
+ \frac{U}{2}\sum_i n_i(n_i-1) - \mu \sum_i n_i,
\]
where \(b_i^\dagger\) and \(b_i\) are bosonic creation and annihilation operators, \(n_i=b_i^\dagger b_i\), \(J\) is the hopping amplitude, \(U\) the on-site repulsion, and \(\mu\) the chemical potential [1405.6941]. Nonequilibrium protocols are implemented through real-time evolution after parameter changes, most prominently interaction quenches \(U_i\to U_f\) at fixed \(J\) and \(\mu\) [1405.6941].

For Markovian open bosonic lattices, the microscopic evolution is given by the Lindblad master equation
\[
\frac{d\rho}{dt} = -i[H,\rho] + \sum_{i,\mu} \left( L_{i\mu}\,\rho\,L_{i\mu}^\dagger - \frac{1}{2}\{L_{i\mu}^\dagger L_{i\mu},\rho\} \right),
\]
with local jump operators \(L_{i\mu}\) and a Bose–Hubbard Hamiltonian scaled as
\[
H = -\frac{J}{z}\sum_{\langle ij\rangle}\left(b_i^\dagger b_j + b_j^\dagger b_i\right) + \sum_i H_{\rm loc}[b_i^\dagger,b_i],
\]
\[
H_{\rm loc} = \omega_0\, b_i^\dagger b_i + \frac{U}{2}\left(b_i^\dagger b_i\right)^2,
\]
where \(z\) is the coordination number and \(n_i=b_i^\dagger b_i\) [2008.02563]. The specific driven-dissipative model studied in the open-system extension uses two-body loss and incoherent pump,
\[
L_{i,2} = \sqrt{\eta}\,b_i^2,\qquad L_{i,1} = \sqrt{f}\,b_i^\dagger,\qquad f=r\,\eta,
\]
with \(\eta\) the two-body loss rate and \(r=f/\eta\) the dimensionless pump-to-loss ratio [2008.02563].

In the rigorous high-dimensional setting, the Bose–Hubbard Hamiltonian is written with explicit \(1/d\) hopping scaling:
\[
H_d \coloneq - \frac{J}{2d} \dGamma(\Delta_d) - \mu \mathcal{N}_\mathcal{F} + \frac{U}{2} \sum_{x\in\Lambda} \mathcal{N}_x (\mathcal{N}_x - 1)
= - \frac{J}{2d}\sum_{\langle x,y\rangle} \big(a^*_x a_y + a^*_y a_x\big) + (J - \mu) \sum_{x\in\Lambda} \mathcal{N}_x + \frac{U}{2} \sum_{x\in\Lambda} \mathcal{N}_x (\mathcal{N}_x - 1),
\]
on the \(d\)-dimensional periodic square lattice \(\Lambda = (\mathbb{Z}/L\mathbb{Z})^d\), with coordination \(z=2d\) [2501.05304]. This scaling is the basis for the large-dimensional limit in which the local mean-field dynamics becomes exact.

## 3. Impurity mapping, Keldysh structure, and self-consistency

In nonequilibrium BDMFT for closed systems, the lattice problem is mapped to an effective local impurity action on the contour \(\mathcal{C}\). In Nambu notation,
\[
\Psi(t) = \begin{pmatrix} b^\dagger(t) \\ b(t) \end{pmatrix},\qquad
\Psi^\dagger(t) = \big(b(t),\; b^\dagger(t)\big),
\]
and the impurity action takes the form
\[
S_{\mathrm{imp}}
=
\int_{\mathcal{C}} dt\;\Big[-\mu(t)\,n(t) + \frac{U}{2} n(t)\big(n(t)-1\big)\Big]
\;+\; \int_{\mathcal{C}} dt\; \Phi_{\mathrm{eff}}^\dagger(t)\,\Psi(t)
\;+\; \frac{1}{2}\int_{\mathcal{C}} dt dt'\;\Psi^\dagger(t)\,\mathbf{\Delta}(t,t')\,\Psi(t'),
\]
where \(\mathbf{\Delta}(t,t')\) is a \(2\times 2\) Nambu hybridization matrix and \(\Phi_{\mathrm{eff}}(t)\) is a self-consistent symmetry-breaking source field [1405.6941]. The source is
\[
\Phi_{\mathrm{eff}}(t) = zJ(t)\,\Phi(t) + \int_{\mathcal{C}} dt'\; \mathbf{\Delta}(t',t)\,\Phi(t'),
\]
with condensate \(\Phi(t)=(\phi^*(t),\phi(t))^T\) and \(\phi(t)=\langle b(t)\rangle\) [1405.6941].

The corresponding contour-ordered Nambu Green’s function is
\[
\mathbf{G}(t,t') = -i\;\langle T_{\mathcal{C}}\,\Psi(t)\,\Psi^\dagger(t')\rangle,
\]
with normal and anomalous entries treated on equal footing [1405.6941]. The impurity Dyson equation is
\[
\big[\mathbf{G}^{-1}\big](t,t') = \big[\mathbf{G}_0^{-1}\big](t,t') - \mathbf{\Sigma}(t,t'),
\]
\[
\big[\mathbf{G}_0^{-1}\big](t,t') = \delta_{\mathcal{C}}(t,t')\;\Big[i\partial_t\,\mathbf{1} + \mu(t)\,\tau_3\Big] - \mathbf{\Delta}(t,t') ,
\]
where \(\mathbf{\Sigma}\) is the local self-energy [1405.6941].

For a Bethe lattice or semi-circular density of states, self-consistency is local:
\[
\mathbf{\Delta}(t,t') = J_\mathrm{eff}^2\;\mathbf{G}_{\mathrm{loc}}(t,t'),
\]
and in the implementation used for a 3D cubic lattice with bandwidth \(W=12J\) and \(z=6\),
\[
\mathbf{\Delta}(t,t') = (3J)^2\;\mathbf{G}_{\mathrm{loc}}(t,t') .
\]
This closure feeds normal and anomalous components back into the impurity problem [1405.6941].

The open-system extension preserves the same basic structure but formulates it directly in Keldysh language for a Lindblad problem. The effective impurity action is
\[
\mathcal{S}_{\rm eff}[b^\dagger_\alpha,b_\alpha]
= S_{\rm loc}[b^\dagger_\alpha,b_\alpha]
+ \int dt\sum_{\alpha=\pm}\alpha\,\boldsymbol{\Phi}_{\rm eff,\alpha}^\dagger(t)\,\boldsymbol{b}_\alpha(t)
- \frac{1}{2}\int dt\,dt'\sum_{\alpha,\beta=\pm}\alpha\beta\,\boldsymbol{b}^\dagger_\alpha(t)\,\boldsymbol{\Delta}^{\alpha\beta}(t,t')\,\boldsymbol{b}_\beta(t'),
\]
with \(\boldsymbol{b}_\alpha=(b_\alpha,\,b^\dagger_\alpha)^T\) in Nambu notation [2008.02563]. The effective field includes a finite-\(z\) memory term,
\[
\boldsymbol{\Phi}^\dagger_{\rm eff,\alpha}(t)
= J\,\boldsymbol{\Phi}^\dagger_{\alpha}(t) + \int dt'\sum_{\beta=\pm}\beta\,\boldsymbol{\Phi}^\dagger_{\beta}(t')\,\boldsymbol{\Delta}^{\beta\alpha}(t',t),
\]
and Bethe-lattice self-consistency reads
\[
\boldsymbol{\Delta}^{\alpha\beta}(t,t') = \frac{J^2}{z}\,\boldsymbol{G}^{\alpha\beta}(t,t') .
\]
The hybridization scales as \(J^2/z\), so the bath vanishes as \(z\to\infty\) and the theory reduces to Gutzwiller mean field [2008.02563].

In the rigorous large-\(d\) treatment, the effective single-site dynamics is instead governed by the nonlinear operator
\[
h^{\varphi} \coloneq - J \big( \alpha_\varphi \, a^\dagger + \overline{\alpha_\varphi} \, a - \abs{\alpha_\varphi}^2 \big) + (J - \mu) \mathcal{N} + \frac{U}{2} \mathcal{N}(\mathcal{N}-1),
\]
with self-consistency through \(\alpha_\varphi = \langle \varphi,a\varphi\rangle\) [2501.05304]. This is explicitly “a mean-field approximation in the hopping amplitude and not in the interaction,” and the interaction remains fully local and nonperturbative [2501.05304].

## 4. Impurity solvers and computational structure

The practical implementation of nonequilibrium BDMFT requires an impurity solver capable of treating strong local interactions and a retarded, self-consistent bath. In the closed-system formulation, the impurity action is solved by a Nambu real-time strong-coupling perturbative approach based on the non-crossing approximation (NCA) [1405.6941]. Introducing pseudo-particle operators \(p_\Gamma^\dagger, p_\Gamma\) for local many-body states \(|\Gamma\rangle\), the NCA pseudo-particle self-energy is
\[
\mathbf{\Sigma}(t,t') = \frac{i}{2}\sum_{\gamma\nu}
\Big\{
\Delta_{\gamma\nu}(t,t')\;[\,b_\gamma^\dagger(t)\,\mathbf{G}(t,t')\,b_\nu(t')\,]
+ \Delta_{\nu\gamma}(t',t)\;[\,b_\gamma(t)\,\mathbf{G}(t,t')\,b_\nu^\dagger(t')\,]
\Big\},
\]
and the pseudo-particle propagators satisfy a contour Dyson equation with cyclic convolution on \(\mathcal{C}\) [1405.6941]. The connected impurity Green’s function is computed from the pseudo-particle bubble,
\[
G_{\gamma\nu}(t,t') = i\,\mathrm{Tr}_\Gamma\big[\mathbf{G}(t',t)\,b_\gamma(t)\,\mathbf{G}(t,t')\,b_\nu(t')\big] + i\,\Phi_\gamma(t)\,\Phi_\nu(t'),
\]
where the second term subtracts the disconnected symmetry-broken contribution [1405.6941].

The open-system impurity solver is also an NCA, but implemented at the super-operator level so that the local Markovian Lindbladian dynamics is treated exactly while the non-Markovian bath is resummed in non-crossing approximation [2008.02563]. The impurity density matrix evolves as
\[
\rho_{\rm imp}(t) = \hat{V}(t,0)\,\rho_{\rm imp}(0),
\]
and the reduced propagator obeys
\[
\hat{V}(t,t') = \hat{V}_0(t,t') + \int_{t'}^t dt_1\int_{t'}^{t_1} dt_2\;\hat{V}_0(t,t_1)\,\hat{S}(t_1,t_2)\,\hat{V}(t_2,t'),
\]
with \(\hat{V}_0\) the bare local Markovian propagator and \(\hat{S}\) the one-particle irreducible self-energy [2008.02563]. At NCA level,
\[
\hat{S}(t_1,t_2) = -\frac{i}{2}\sum_{\alpha\beta}\sum_{a,b}\alpha\beta
\left[
\Delta^{\beta\alpha}_{b a}(t_1,t_2)\;\hat{b}^\dagger_{\beta b}\;\hat{V}(t_1,t_2)\;\hat{b}_{\alpha a}
+ \Delta^{\alpha\beta}_{a b}(t_2,t_1)\;\hat{b}_{\beta b}\;\hat{V}(t_1,t_2)\;\hat{b}^\dagger_{\alpha a}
\right],
\]
where \(\hat{b}_\pm\) act by left or right multiplication on the density matrix [2008.02563]. In stationary problems, time-translation invariance simplifies the solver because \(\boldsymbol{\Delta}(t,t')=\boldsymbol{\Delta}(t-t')\) and \(\hat{V}(t,t')=\hat{V}(t-t')\) [2008.02563].

A compact summary of the principal computational ingredients is useful.

| Formulation | Effective local object | Solver structure |
|---|---|---|
| Closed-system nonequilibrium BDMFT | Contour impurity action with \(\mathbf{\Delta}(t,t')\) and \(\Phi_{\mathrm{eff}}(t)\) | Nambu real-time strong-coupling NCA [1405.6941] |
| Open-system nonequilibrium BDMFT | Lindbladian impurity with coherent field and non-Markovian bath | Super-operator hybridization expansion with NCA [2008.02563] |
| High-dimensional Weiss-field limit | Nonlinear single-site Schrödinger equation \(i\partial_t\varphi=h^\varphi\varphi\) | Direct integration after Hilbert-space truncation [2501.05304] |

In numerical practice, the closed-system implementation uses a uniform time grid on \(\mathcal{C}\) and a fifth-order multi-step method; the open-system stationary solver uses an implicit second-order Runge–Kutta scheme with typical parameters \(dt\approx 0.004\), \(t_{\rm max}\approx 10\), and local Hilbert-space truncation \(\mathrm{dim}_H\approx 10\)–14 [1405.6941, 2008.02563]. The rigorous high-dimensional work proves global well-posedness of the truncated single-site problem and then controls the \(M\to\infty\) limit [2501.05304].

## 5. Dynamical regimes in closed bosonic lattices

Applied to interaction quenches in the Bose–Hubbard model, nonequilibrium BDMFT captures several dynamical regimes that are absent or frozen in Gutzwiller mean-field theory because the latter lacks the retarded hybridization \(\mathbf{\Delta}(t,t')\) and therefore cannot describe hopping-induced relaxation, damping, or thermalization at finite temperature [1405.6941].

For quenches starting from a normal or Mott state with \(\phi=0\), the theory identifies rapid thermalization near intermediate \(U_f\) and prethermalization plateaus at both weak and strong \(U_f\) [1405.6941]. The prethermal behavior is attributed to proximity to the integrable limits \(U=0\) and \(U\to\infty\), while a qualitative non-equilibrium phase diagram in the \((U_f,T_{\mathrm{eff}})\) plane delineates the intermediate region of fast thermalization [1405.6941]. Thermalization in the symmetric phase is monitored by
\[
\kappa(t) = \frac{\langle n^2\rangle(t) - \langle n^2\rangle_{U_i,T_i}}{\langle n^2\rangle_{U_f,T_{\mathrm{eff}}} - \langle n^2\rangle_{U_i,T_i}},
\]
with \(\kappa(0)=0\) and \(\kappa=1\) indicating thermalization [1405.6941].

For quenches starting from a superfluid initial state, five regimes are reported: deep-Mott collapse-and-revival oscillations with frequency \(\omega\approx U_f\) and exponential damping; intermediate-coupling relaxation to the normal phase after a few oscillations; trapping in a nonthermal superfluid near the phase boundary on the normal side; excitation of a coherent amplitude mode with roughly constant frequency and damping; and rapid self-amplified growth of the condensate for small quenches that remain within the equilibrium superfluid region [1405.6941]. Near the dynamical transition \(U_f=c(U_i,T_i)\), the theory finds a kink in the phase frequency \(\omega_\theta\), a maximal relaxation rate for double occupancy, and maximal damping of the amplitude mode [1405.6941].

Energy conversion is an explicit diagnostic. The kinetic, condensate, and interaction contributions are
\[
E_k(t) = \frac{i}{2}\,\mathrm{Tr}\big[(\mathbf{\Delta}*\mathbf{G})^<(t,t)\big],\qquad
E_c(t) = -zJ(t)\,|\phi(t)|^2,\qquad
E_i(t) = \frac{U(t)}{2}\,\big(\langle n^2\rangle(t)-\langle n\rangle(t)\big),
\]
and nonequilibrium BDMFT shows nontrivial exchange between these channels during relaxation, in contrast to Gutzwiller mean-field in symmetric states [1405.6941].

Finite temperature plays a systematic role: increasing the initial temperature \(T_i\) enhances damping, accelerates relaxation, suppresses the plateau values of \(\bar{|\phi|}\), and reduces the trapped superfluid regions [1405.6941]. This establishes nonequilibrium BDMFT as a finite-temperature dynamical theory rather than merely a zero-temperature strong-coupling approximation.

## 6. Open-system extension: dissipation, finite-frequency instability, and synchronization

In driven-dissipative bosonic lattices, nonequilibrium BDMFT extends beyond coherent relaxation and addresses steady states of Lindbladian many-body systems [2008.02563]. The open-system theory reveals qualitative phenomena that are completely missed by Gutzwiller mean-field because, in the normal phase with \(\boldsymbol{\Phi}=0\), Gutzwiller predicts uncoupled sites and therefore no dependence of local stationary properties on the hopping \(J\) [2008.02563]. DMFT restores this dependence through the non-Markovian bath \(\boldsymbol{\Delta}\), which encodes virtual hopping processes and finite-connectivity memory.

In the normal phase, the retarded and Keldysh hybridizations have distinct roles:
\[
\mathcal{S}_{\rm eff}
= S_{\rm loc}[b^\dagger_{\rm cl/q},b_{\rm cl/q}]
+ \int dt\,\boldsymbol{\Phi}^\dagger_{\rm eff,cl}(t)\,\boldsymbol{b}_{\rm q}(t)
- \frac{1}{2}\int dt\,dt'
\Big[\boldsymbol{b}^\dagger_{\rm q}(t)\,\boldsymbol{\Delta}^R(t,t')\,\boldsymbol{b}_{\rm cl}(t') + \mathrm{h.c.}\Big]
- \frac{1}{2}\int dt\,dt'\,\boldsymbol{b}^\dagger_{\rm q}(t)\,\boldsymbol{\Delta}^K(t,t')\,\boldsymbol{b}_{\rm q}(t'),
\]
where \(\Delta^R\) encodes spectral damping and \(\Delta^K\) bath occupation or noise [2008.02563]. Because the system is far from equilibrium, \(\Delta^R\) and \(\Delta^K\) are independent and are not constrained by an FDT relation [2008.02563].

The theory identifies hopping-induced dissipative channels in which a boson is pumped on one site, hops to a neighbor, and is lost there through the local dissipators [2008.02563]. These channels suppress local gain, redistribute steady-state populations, and reduce population inversion [2008.02563]. At large two-body loss \(\eta\), the steady state enters a stationary quantum-Zeno regime controlled by the effective scale
\[
\Gamma_{\rm eff} = \frac{2(J/z)^2}{U^2+\eta^2}\,\eta,
\]
with non-monotonic dependence on \(\eta/U\) and a collapse of observables as functions of \(\Gamma_{\rm eff}\) [2008.02563]. In this regime, probabilities for \(n\ge 2\) become exponentially suppressed and the steady-state weight concentrates in the \(n=0,1\) sector [2008.02563].

The open-system phase structure contains a normal phase with \(\langle b_i\rangle=0\) and a non-equilibrium superfluid with a time-oscillating order parameter
\[
\psi(t)=\langle b_i(t)\rangle \propto e^{-i\Omega_c t},
\]
which breaks both \(U(1)\) and time-translation invariance [2008.02563]. In the normal phase, the spectral function
\[
A(\omega)=-(1/\pi)\mathrm{Im}\,G^R(\omega)
\]
may develop a negative-density-of-states region \(A(\omega)<0\) for \(0<\omega<\Omega_0(r,J)\), implying gain and negative absorbed power under a weak coherent probe through
\[
\dot{W}=v_0^2\,\omega A(\omega) .
\]
The onset of the ordered phase is a finite-frequency instability determined, on the Bethe lattice, by
\[
\mathrm{Im}\,G^R(\Omega_c,J_c) = 0,
\]
\[
\frac{1}{J_c} + \mathrm{Re}\,G^R(\Omega_c,J_c) + \frac{J_c}{z}\left[\mathrm{Re}\,G^R(\Omega_c,J_c)\right]^2 = 0.
\]
This instability has no equilibrium analog because in equilibrium \(\mathrm{Im}\,G^R(0)=0\) follows from FDT, whereas here the unstable mode appears at a nonzero \(\Omega_c\) [2008.02563].

Finite-connectivity fluctuations strongly renormalize the phase boundary relative to Gutzwiller mean field. The ordered region shrinks as \(z\) decreases and moves toward larger \(J\) and \(r\) because the DMFT bath-induced dissipation suppresses negative density of states and local gain [2008.02563]. The same transition is connected to many-body synchronization of quantum van der Pol oscillators. In the semiclassical limit, the Keldysh action yields
\[
i\dot{b}_{j,{\rm cl}}
=
\left[\omega_0 + \frac{U}{2}|b_{j,{\rm cl}}|^2\right]\,b_{j,{\rm cl}}
+ i\left[\frac{f}{2} - \frac{\eta}{2}|b_{j,{\rm cl}}|^2\right]\,b_{j,{\rm cl}}
+ \frac{J}{z}\sum_{\langle j'\rangle} b_{j',{\rm cl}} + \xi_j(t),
\]
with white noise \(\langle \xi_i(t)\xi_j(t')\rangle = f\,\delta_{ij}\delta(t-t')\), so the finite-frequency superfluid instability becomes the onset of a collective limit cycle [2008.02563].

## 7. Large-dimensional limit, rigor, and relation to mean-field theory

A rigorous mathematical foundation for the simplest nonequilibrium bosonic DMFT closure is provided by the high-dimensional analysis of the Bose–Hubbard model with hopping scaled as \(J/(2d)\) [2501.05304]. The central result is that the one-site reduced density matrix of the exact Schrödinger evolution converges, in trace norm, to a rank-one projector \(p_\varphi(t)=|\varphi(t)\rangle\langle\varphi(t)|\) generated by the nonlinear single-site mean-field equation [2501.05304].

The exact reduced one-site dynamics satisfies
\[
i \partial_t \gamma_d^{(1)}
= [h^{\varphi}, \gamma_d^{(1)}]
- J \, \mathrm{PTr}_{2}\Big[ \big((a^* - \overline{\alpha}) \otimes (a - \alpha) + (a - \alpha)\otimes(a^* - \overline{\alpha})\big), \gamma_d^{(2)}\Big],
\]
so the deviation from the mean-field closure is controlled by the two-site reduced density matrix \(\gamma_d^{(2)}\) [2501.05304]. The large-\(d\) analysis shows that these corrections vanish with increasing dimension.

Two explicit error bounds are established. Under exponential-tail assumptions on the initial occupation distribution, Theorem 1 gives a trace-norm estimate whose \(d\)-dependent prefactor tends to zero as \(d\to\infty\) [2501.05304]. Under the assumptions \(U>0\) and \(\mathrm{Tr}(p_\varphi(0)\mathcal{N}^4)\le C\), Theorem 2 yields
\[
\| \gamma_d^{(1)}(t) - p_{\varphi}(t)\|_{\mathcal{L}^1} \leq C(J,\mu,U)\, e^{C(J,\mu,U) (1+t^7)} \left( \mathrm{Tr}\big[\gamma_d^{(1)}(0) \big( q_\varphi(0) \mathcal{N} ^2 q_\varphi(0) +q_\varphi(0)\big)\big] +\frac{1}{d} \right)^{1/2},
\]
and for initial Gutzwiller product states this reduces to
\[
\| \gamma_d^{(1)}(t) - p_{\varphi}(t)\|_{\mathcal{L}^1} \leq C(J,\mu,U)\, e^{C(J,\mu,U) (1+t^7)} \frac{1}{\sqrt{d}} .
\]
The same work proves convergence of the order parameter,
\[
\alpha_{\mathrm{micro}}(t) := \frac{1}{|\Lambda|} \sum_{x \in \Lambda} \langle \Psi_d(t), a_x \Psi_d(t) \rangle \to \alpha_{\varphi(t)} := \langle \varphi(t),a \varphi(t) \rangle \qquad \text{as } d\to\infty .
\]
This rigorously justifies the time-dependent Weiss field as the large-dimensional limit of the microscopic condensate [2501.05304].

The rigorous results also clarify the relation between BDMFT and Gutzwiller theory. The proven equation corresponds to the time-dependent Gutzwiller or Weiss-field dynamics, not to the full hybridization-based Keldysh impurity theory [2501.05304]. A plausible implication is that the rigorous large-\(d\) limit establishes the controlled starting point from which full nonequilibrium BDMFT incorporates finite-connectivity corrections through the dynamical bath \(\Delta(t,t')\), as made explicit in the closed- and open-system DMFT constructions [1405.6941, 2008.02563].

## 8. Comparison with related approaches, limitations, and open directions

Compared with Gutzwiller mean-field theory, nonequilibrium BDMFT retains local interactions nonperturbatively while adding a retarded hybridization term that captures memory, damping, and thermalization [1405.6941]. In the open-system context, this same retarded bath produces hopping-induced dissipative channels and finite-\(z\) memory feedback that are absent in Gutzwiller mean field, especially in the normal phase where Gutzwiller predicts completely uncoupled sites [2008.02563]. The distinction is therefore not merely quantitative.

Compared with equilibrium or low-frequency strong-coupling approaches, nonequilibrium BDMFT is formulated directly in real time and at finite temperature, with full Nambu structure and explicit treatment of normal and anomalous components [1405.6941]. Compared with fermionic DMFT, the bosonic problem requires explicit handling of condensation and anomalous propagators; in the open-system case it additionally requires a Lindbladian Keldysh formalism in which \(\Delta^R\) and \(\Delta^K\) are independent and no FDT constraint exists [2008.02563].

The theory also has clear limitations. DMFT neglects nonlocal spatial correlations and is controlled in the limit of large coordination or high dimension [1405.6941, 2008.02563, 2501.05304]. The NCA impurity solvers are approximate: in the closed-system case spectral sum-rule checks show small errors deep in the Mott regime but larger ones near the superfluid boundary; in the open-system case higher-order crossing diagrams may be needed for very small \(z\) or deep inside symmetry-broken phases [1405.6941, 2008.02563]. Numerical cost is substantial because real-time two-time objects scale as \(N_t^2\) in memory and bosonic local Hilbert spaces grow rapidly with occupancy [1405.6941]. In the open-system steady-state solver, accessible parameter ranges are also limited by Hilbert-space truncation, particularly at high pump \(r\) [2008.02563].

Several extensions are explicitly identified. For driven-dissipative bosons, possible developments include cluster DMFT for short-range correlations and improved impurity solvers such as full hybridization expansion via diagrammatic Monte Carlo, auxiliary master equations with a discretized bath, MPO-based non-Markovian impurity solvers, and renormalized Lindblad NRG [2008.02563]. From the rigorous side, extending the analysis from the Weiss-field closure to full nonequilibrium bosonic DMFT with anomalous Nambu components and a dynamic hybridization bath remains an open problem [2501.05304].

Taken together, these results define nonequilibrium bosonic dynamical mean-field theory as a hierarchy of local self-consistent descriptions for bosonic lattice dynamics: at the simplest level, a rigorously justified Weiss-field evolution in large dimension; at the full DMFT level, a Keldysh impurity theory with retarded hybridization that captures relaxation, damping, thermalization, dissipative renormalization, finite-frequency instabilities, and collective synchronization in both closed and Markovian open quantum many-body systems [1405.6941, 2008.02563, 2501.05304].

Source: https://www.emergentmind.com/topics/nonequilibrium-bosonic-dynamical-mean-field-theory