---
title: Open-System Keldysh–Lindblad Framework
url: https://www.emergentmind.com/topics/open-system-keldysh-lindblad-framework
type: topic
---

# Open-System Keldysh–Lindblad Framework

Searching arXiv for recent and foundational papers on the open-system Keldysh–Lindblad framework.
The open-system Keldysh-Lindblad framework is a closed-time-path formulation of open quantum dynamics in which the Markovian evolution of a reduced density matrix, written in Lindblad form, is recast as a Schwinger-Keldysh functional integral over forward and backward branches, or, in an equivalent operator-language construction, as dynamics in Liouville space under third quantization. In this framework, coherent Hamiltonian evolution, dissipation, and noise appear in a unified representation that makes transparent the relation between Lindblad master equations, Keldysh actions, and, for bosons, phase-space objects such as the Wigner and characteristic functions [1512.00637] [2302.14047].

## 1. Lindblad dynamics as the starting point

The framework begins from the generic Markovian master equation
\[
\partial_t \rho \;=\; -\,i\,[H,\rho] \;+\;\sum_\alpha
\gamma_\alpha\Bigl(L_\alpha\,\rho\,L_\alpha^\dagger
\;-\;\tfrac12\{L_\alpha^\dagger L_\alpha,\rho\}\Bigr),
\]
or, equivalently,
\[
\partial_t\rho \;=\;\mathcal L[\rho]
\;=\;
-\,i\,[H,\rho]
\;+\;\sum_j\gamma_j\Bigl(L_j\,\rho\,L_j^\dagger\;-\;\tfrac12\{L_j^\dagger L_j,\rho\}\Bigr),
\]
with \(H\) at most quadratic in the creation/annihilation operators in the quadratic setting, \(L_j\) linear in those same modes, and \(\gamma_j\ge 0\) [1512.00637] [2302.14047].

This Liouvillian formulation encodes the reduced dynamics of an open system after environmental degrees of freedom have been eliminated. In the field-theoretical derivation based on a total Hamiltonian \(H=H_S+H_B+H_{SB}\) with \(H_{SB}=\sum_\alpha S^\alpha B^\alpha\), the reduced density matrix is evolved by a transmission matrix on a closed contour. Grouping bath contractions into irreducible blocks yields a Dyson equation for that transmission matrix, and differentiating it gives an exact non-Markovian master equation,
\[
\frac{d}{dt}\rho_S(t)
= -\,i\,[H_S,\rho_S(t)]
+ \int_{t_i}^t dt'\,\Sigma(t,t')\,\rho_S(t'),
\]
where \(\Sigma(t,t')\) is an operator super-kernel [2202.05203].

The relation between the exact non-Markovian equation and the Lindblad-GKSL equation is not assumed but derived. In the Born approximation, the irreducible self-energy is determined by bath Green functions \(D^{\alpha\beta}(t,t')\). A quasiparticle or pole approximation, described as equivalent to the usual assumption of a timescale separation, makes the kernel frequency independent; imposing the rotating-wave approximation then yields a Markov master equation of Lindblad form, with jump operators, rates from the bath spectral density, and a Lamb shift [2202.05203]. This establishes the Lindblad equation not as an isolated ansatz but as a controlled limit within a broader contour-based formalism.

## 2. Closed-time contour, coherent states, and Keldysh rotation

The functional-integral version of the framework uses a closed time contour with a forward \((+)\) and backward \((-)\) branch. The density matrix is evolved in infinitesimal time steps, coherent-state resolutions of identity are inserted on each branch, and operator monomials are replaced by their coherent-state symbols. In the continuum limit, this yields the closed-time-path functional integral
\[
Z
= \int\!\mathscr{D}[\psi_+,\psi_-]\,
e^{\,iS[\psi_+,\psi_-]}
=1,
\]
with action
\[
S \;=\;
\int_{-\infty}^{\infty}dt\;\Bigl[\,
\psi_+^*\,i\partial_t\psi_+
\;-\;\psi_-^*\,i\partial_t\psi_-
\;-\;i\,\mathcal{L}(\psi_+^*,\psi_+;\psi_-^*,\psi_-)
\Bigr].
\]
The super-Liouvillian symbol is
\[
\mathcal{L}(\psi_+^*,\psi_+;\psi_-^*,\psi_-)
= -\,i\bigl[H(\psi_+^*,\psi_+)-H(\psi_-^*,\psi_-)\bigr]
+\sum_\alpha\gamma_\alpha\Bigl(L_{\alpha,+}L_{\alpha,-}^*
\;-\;\tfrac12[L_{\alpha,+}^*L_{\alpha,+}+L_{\alpha,-}^*L_{\alpha,-}]\Bigr)
\]
[1512.00637].

The standard Keldysh rotation introduces classical and quantum fields,
\[
\phi_c = \tfrac1{\sqrt2}\,(\psi_+ + \psi_-),\qquad
\phi_q = \tfrac1{\sqrt2}\,(\psi_+ - \psi_-),
\]
after which the quadratic part of the action takes the canonical structure
\[
S
= \int_{t,\mathbf x}\!\bigl[
\phi_q^*\,D^R\,\phi_c
+ \phi_c^*\,D^A\,\phi_q
+ \phi_q^*\,D^K\,\phi_q
+ S_{\rm int}[\phi_c,\phi_q]
\bigr].
\]
Here \(D^{R/A/K}\) are the inverse retarded, advanced, and Keldysh propagators; \(S_{\rm int}\) contains non-quadratic interaction and noise terms [1512.00637].

This decomposition is technically central because it separates response structure from fluctuations. In the free damped-oscillator example,
\[
D^R(\omega)=\omega-\omega_0+i\kappa,\quad
D^A(\omega)=D^R(\omega)^*,\quad
D^K=2i\kappa,
\]
and
\[
G^K(\omega)=-G^R\,D^K\,G^A
=-\tfrac{2i\kappa}{(\omega-\omega_0)^2+\kappa^2}.
\]
In thermal equilibrium, replacing \(D^K\to i2\delta(2n(\omega)+1)\) recovers the familiar quantum fluctuation-dissipation relation
\[
G^K(\omega)=(2n(\omega)+1)\bigl[G^R(\omega)-G^A(\omega)\bigr].
\]
The same formal machinery therefore accommodates both thermal and genuinely nonthermal driven-dissipative settings [1512.00637].

## 3. Third quantization, Liouville-space supermodes, and dissipative symmetry

A more explicit link between Lindblad and Keldysh formulations is obtained by vectorizing the density matrix, \(\rho\to|\rho\>\!\>\), in Liouville space and introducing superoperators. For bosons, McDonald and Clerk define
\[
\begin{aligned}
\hat a_{\,m,\rm cl}\,|\rho\>\!\>&\equiv\frac1{\sqrt2}\,\bigl|\{\hat a_m,\rho\}\bigr\>\!\>, \\
\hat a_{\,m,\rm q}\,|\rho\>\!\>&\equiv\frac1{\sqrt2}\,\bigl|[\hat a_m,\rho]\bigr\>\!\>,\\
\hat a^\dagger_{\,m,\rm cl}\,|\rho\>\!\>&\equiv\frac1{\sqrt2}\,\bigl|\{\hat a_m^\dagger,\rho\}\bigr\>\!\>, \\
\hat a^\dagger_{\,m,\rm q}\,|\rho\>\!\>&\equiv\frac1{\sqrt2}\,\bigl|[\hat a_m^\dagger,\rho]\bigr\>\!\>,
\end{aligned}
\]
with only nonzero commutators
\[
[\,\hat a_{m,\rm cl}\,,\,\hat a^\dagger_{n,\rm q}\,]
\;=\;
[\,\hat a_{m,\rm q}\,,\,-\hat a^\dagger_{n,\rm cl}\,]
\;=\;\delta_{mn}\,\hat 1_{\rm sup}.
\]
For fermions, a parity operator is inserted so that the resulting super-modes obey canonical anticommutation relations [2302.14047].

In the bosonic single-mode example, the Lindbladian becomes
\[
\hat{\mathcal L}
=
(\omega_0 - i\,\tfrac\kappa2)\,\hat a^\dagger_{\rm q}\,\hat a_{\rm cl}
\;+\;
(\omega_0 + i\,\tfrac\kappa2)\,\hat a_{\rm q}\,\hat a^\dagger_{\rm cl}
\;-\;
i\,\kappa\,(2\bar n+1)\,\hat a^\dagger_{\rm q}\,\hat a_{\rm q}.
\]
By inserting coherent-state resolutions of identity for the supermodes, one obtains a finite-time coherent-state path integral whose action is the standard quadratic Keldysh action. In compact notation,
\[
S
=\int dt\;\Bigl[
a_q^\dagger\,(i\partial_t-H_{\rm eff})\,a_c
\;+\;
a_c^\dagger\,(i\partial_t-H_{\rm eff}^\dagger)\,a_q
\;+\;
i\,a_q^\dagger\,\Gamma\,a_q
\Bigr],
\]
with \(H_{\rm eff}=\omega_0-i\kappa/2\) and \(\Gamma=\kappa(2\bar n+1)\). Multi-mode bosonic or fermionic generalizations replace \(a_{c/q}\) by vectors \(\phi_{c/q}\), \(H_{\rm eff}\) by a non-Hermitian dynamical matrix, and \(\Gamma\) by the noise correlator [2302.14047].

A central structural result is the dissipative symmetry of quadratic Lindbladians. There is no pure \(\hat a^\dagger_{\rm cl}\hat a_{\rm cl}\) term; trace-preservation forbids a classical-classical block. Moreover, the Keldysh or noise term \(\hat a^\dagger_{\rm q}\hat a_{\rm q}\) can be exactly removed by the similarity transformation
\[
\mathcal V=e^{-\,(2\bar n+1)\,\hat a^\dagger_{\rm q}\hat a_{\rm q}},
\]
which sends
\[
\mathcal V^{-1}\hat{\mathcal L}\mathcal V
=(\omega_0-i\tfrac\kappa2)\,\hat a^\dagger_{\rm q}\hat a_{\rm cl}
+(\omega_0+i\tfrac\kappa2)\,\hat a_{\rm q}\hat a^\dagger_{\rm cl}.
\]
The remaining bilinear can then be diagonalized by introducing non-Hermitian quasiparticles \(\alpha,\beta\), and the full Lindbladian spectrum and eigen-supermodes follow much as for a quadratic Hamiltonian. In Keldysh language, the same symmetry appears as the statement that the retarded and advanced blocks are unaffected by \(\bar n\), while the Keldysh block alone carries all of the noise. For bosons, the Wigner function and the characteristic function may be viewed as wavefunctions of the density matrix in the eigenbasis of these superoperators [2302.14047].

## 4. From exact contour dynamics to Markovian Keldysh-Lindblad theory

The contour formalism also supports a systematic derivation of reduced dynamics before the Markov approximation is taken. The transmission super-operator is written as
\[
\rho_S(t)=T(t,t_i)\rho_S(t_i),
\]
with
\[
T(t,t_i)
= \mathrm{Tr}_B\!\left\{T_C\exp\!\left[ -\,i\!\int_C\!d\tau\,S^\alpha(\tau)B^\alpha(\tau) \right]\rho_B\right\},
\]
where \(T_C\) denotes contour ordering. In the absence of coupling, one has
\[
T_0(t,t') = G_R(t,t')\otimes G_A(t',t),
\]
with
\[
G_R(t,t') = -\,i\,\theta(t-t')\,e^{-iH_S(t-t')},\qquad
G_A(t',t) = +\,i\,\theta(t-t')\,e^{-iH_S(t-t')}.
\]
Irreducible bath contractions define a self-energy \(\Sigma(\tau_1,\tau_2)\), and reconnecting them with \(T_0\) gives a matrix Dyson equation on the contour [2202.05203].

For a Caldeira-Leggett bath in thermal equilibrium, with \(H_B=\sum_k\Omega_k b_k^\dagger b_k\), \(B^1=B=\sum_k\lambda_k b_k\), and \(B^2=B^\dagger\), the nonzero bath correlators are
\[
D^{12}(t,t')=\sum_k\lambda_k^2[1+n_k]e^{-i\Omega_k(t-t')},\qquad
D^{21}(t,t')=\sum_k\lambda_k^2 n_k e^{+i\Omega_k(t-t')},
\]
where \(n_k=(e^{\beta\Omega_k}-1)^{-1}\). To second order in \(H_{SB}\), the irreducible self-energy is the sum of four terms built from \(S^\alpha\), \(G_{R/A}\), and \(D^{\alpha\beta}\) [2202.05203].

The Markovian limit is then formulated in frequency space. The poles of \(T(\omega)\), determined by \(\det\{[T_0]^{-1}-\Sigma\}=0\), fix the characteristic frequencies. In lowest weak-coupling order one replaces \(\Sigma(\omega)\) by \(\Sigma(E_p-E_{p'})\); this is identified as the Markov or timescale-separation approximation. Inverse Fourier transforming gives a time-local kernel \(\Sigma_M\), and decomposing it into Hermitian and anti-Hermitian parts yields the Lindblad-GKSL form with \(H_{\rm Lamb}\) and dissipative rates \(\gamma_{\alpha\beta}(\omega)\) [2202.05203].

The single-qubit example makes the logic explicit. For
\[
H_S=(\omega_0/2)\sigma^z,\qquad
H_{SB}=\sigma^+B+\sigma^-B^\dagger,
\]
the Markov-RWA limit gives
\[
\gamma_{\downarrow}=J(\omega_0)[1+n(\omega_0)],\qquad
\gamma_{\uparrow}=J(\omega_0)n(\omega_0),
\]
and the master equation
\[
\frac{d\rho}{dt}
= -i[(\omega_0/2+\Delta)\sigma^z,\rho]
+\gamma_{\downarrow}(\sigma^-\rho\sigma^+ -\tfrac12\{\sigma^+\sigma^-,\rho\})
+\gamma_{\uparrow}(\sigma^+\rho\sigma^- -\tfrac12\{\sigma^-\sigma^+,\rho\}),
\]
with \(\Delta=P\int d\omega\,J(\omega)/(\omega_0-\omega)\). A common misconception is that Keldysh field theory and Lindblad dynamics are separate techniques. The contour derivation shows instead that the Lindblad description arises as a specific approximation within the contour-Dyson framework [2202.05203].

## 5. Many-body field theory, nonthermal criticality, and fluctuation structure

In many-body driven-dissipative problems, the Keldysh-Lindblad framework is used directly as a non-equilibrium field theory. A minimal bosonic model with pumping and loss is specified by
\[
H=\int d^d x\;\bigl[\psi^\dagger(-K_c\nabla^2+r_c)\psi+u_c|\psi|^4\bigr],\qquad
\mathcal L\,\rho=\int d^d x\;\bigl[\gamma_p\mathcal D[\psi^\dagger]+\gamma_\ell\mathcal D[\psi]+u_d\mathcal D[\psi^2]\bigr].
\]
Its Keldysh action reads
\[
S
=\int_{t,\mathbf x}\Bigl\{
\phi_q^*\bigl[i\partial_t+K_c\nabla^2-r_c+i r_d\bigr]\phi_c
-\bigl(u_c-i u_d\bigr)\bigl(\phi_q^*\phi_c^*\phi_c^2+c.c.\bigr)
+i2\gamma\,\phi_q^*\phi_q
\Bigr\},
\]
with \(r_d=(\gamma_\ell-\gamma_p)/2\) and \(\gamma=(\gamma_\ell+\gamma_p)/2\). The saddle-point equations give \(\phi_q=0\), and above threshold \(r_d<0\) one finds \(|\phi_c|^2=\rho_0=-r_d/u_d\). Expanding about this condensate yields a quadratic Nambu-Keldysh action whose retarded inverse propagator has a complex Bogoliubov spectrum; in the limit \(q\to 0\), one finds a diffusive Goldstone mode \(\omega\sim -iDq^2\) [1512.00637].

Near the condensation critical point, power counting shows that only \(\phi_q^*\phi_c^*\phi_c^2\) and the noise term \(\phi_q^*\phi_q\) survive as marginal couplings in \(d\le 3\). The resulting Langevin equation is
\[
i\partial_t\psi
=\Bigl[-(K_c-iK_d)\nabla^2 + r_c- i r_d +(u_c-i u_d)|\psi|^2\Bigr]\psi
+\xi,
\]
with
\[
\langle\xi(t)\xi^*(t')\rangle
=2\gamma\,\delta(t-t')\delta(\mathbf x-\mathbf x').
\]
In the ordered phase, the phase-only reduction gives the KPZ action
\[
S_{\rm KPZ}=\int dt\,d^d x\;\tilde\theta
\Bigl[\partial_t\theta -D\nabla^2\theta
-\tfrac{\lambda}{2}(\nabla\theta)^2 -\Delta\,\tilde\theta\Bigr],
\]
and in \(d=1,2\) this implies non-equilibrium KPZ scaling of correlations. Functional RG treatments supplement the action by an infrared regulator and use the Wetterich equation; in \(d=3\), one finds an IR-attractive equilibrium fixed point with purely dissipative dynamics, together with a distinct universal drive-decay exponent \(\eta_r\neq 0\) controlling how quickly coherent couplings vanish at the longest scales [1512.00637].

A different line of development formulates the Lindblad contribution itself as a double-trace deformation. For a real scalar field, locality and Poincaré invariance lead to
\[
S[\phi_+,\phi_-]
= \int d^dx\Bigl\{L_0(\phi_+) - L_0(\phi_-) - i\sum_i\gamma_i\bigl[O_{i,+}O_{i,-}^*-\tfrac12 O_{i,+}^*O_{i,+}-\tfrac12 O_{i,-}^*O_{i,-}\bigr]\Bigr\},
\]
with \(\gamma_i\ge 0\). In the Keldysh basis this becomes
\[
S=S_0[\phi_c,\phi_q]+i\gamma\int O_q^2,
\]
showing explicitly the dissipative deformation \(i\gamma O_q^2\). If \(O_q\) has scaling dimension \(\Delta\), then \([\gamma]=d-2\Delta\), and at leading order in \(1/N\),
\[
G_f(k)=[G_0(k)^{-1}+i\gamma]^{-1}.
\]
The beta function
\[
\beta(\gamma)=\frac{d\gamma}{d\ln\mu}=(d-2\Delta)\gamma+O(\gamma^2/N)
\]
implies that dissipation is IR-relevant for \(\Delta<d/2\) and UV-relevant for \(\Delta>d/2\). The resulting fixed points are nonunitary and nonthermal, and the limiting Keldysh component violates any fluctuation-dissipation relation of the form
\[
G^K \propto (G^R-G^A)\coth(\beta\omega/2).
\]
This directly addresses the common assumption that dissipation effects are always IR relevant: the framework presented in "Double-trace deformation in Keldysh field theory" states that Weinbergian constraints do not forbid a UV-relevant dissipative operator in a driven-dissipative setting [2012.05379].

## 6. Weak nonlinearities, exact treatments, and lattice real-time simulation

The framework is not restricted to purely quadratic problems. In the third-quantized construction, any quadratic or weakly non-linear Lindbladian can be mapped to a Keldysh action by writing \(\mathcal L\) in the \(\{\hat a_{\rm cl},\hat a_{\rm q}\}\) basis, introducing coherent-state resolutions of identity for the supermodes, and reading off the retarded/advanced kernels \(i\partial_t-H_{\rm eff}\) and the noise kernel \(\Gamma\). For weak nonlinearities, such as a Kerr term \(\tfrac U2[\hat a^\dagger\hat a^\dagger\hat a\hat a,\rho]\), one adds quartic vertices
\[
U\,a_q^\dagger a_c\,(a_c^\dagger a_c +a_q^\dagger a_q)
\]
to the action. A further gauge transform, which shifts the phase by the time-integral of the amplitude, can often render that vertex quadratic at the expense of a phase factor in the source terms; one then performs a Gaussian integral mode-by-mode and reconstructs closed-form expressions for correlation functions or even the full Wigner function [2302.14047].

A distinct computational realization appears in lattice Schwinger-Keldysh simulations of Lindblad dynamics for non-relativistic spinless fermions. The generating functional includes two real-time branches and an imaginary-time leg enforcing a thermal initial condition,
\[
Z=\int D\psi^*D\psi\;\exp\Bigl[iS_{\rm real}[\psi_+,\psi_-]-S_{\rm imag}[\psi_0]\Bigr].
\]
On a three-dimensional cubic lattice, the hopping Hamiltonians are
\[
H_\pm=\sum_{x,k}\Bigl[\bigl(-w+iA_k(x)\bigr)\psi_\pm^\dagger(x)\psi_\pm(x+e_k)
+\bigl(-w-iA_k(x)\bigr)\psi_\pm^\dagger(x+e_k)\psi_\pm(x)\Bigr],
\]
while the jump operators are Hermitian current densities
\[
L_{k\pm}(x)=j_{k\pm}(x)
=i\Bigl[\psi_\pm^\dagger(x)\psi_\pm(x+e_k)-\psi_\pm^\dagger(x+e_k)\psi_\pm(x)\Bigr].
\]
For isotropic dissipation \(\gamma_k=\gamma\), the dissipator is rewritten as
\[
i\,\frac{\gamma}{2}\sum_k\bigl(j_{k+}-j_{k-}\bigr)^2,
\]
which admits a Hubbard-Stratonovich decoupling by real Gaussian fields \(B_k(x,t)\) [2111.04937].

After decoupling, the fermions can be integrated out exactly:
\[
Z
= \int DB\;\det D[B]\;\exp\Bigl[-\tfrac1{2\gamma}\int dt\sum_{x,k}B_k^2\Bigr].
\]
A backward-difference discretization makes \(D\) a block matrix, and when the number of imaginary-time slices \(N_\tau\) is even, one can pair factors so that the determinant is nonnegative. In the continuum limit,
\[
\det D=\det[1+e^{-\beta h_0}]+\mathcal O(\delta^2),
\]
which is manifestly independent of the noises \(B_k\). The determinant can therefore be dropped without approximation, leaving a positive Gaussian measure for \(B_k\). In Keldysh space, one defines
\[
\psi_c=\tfrac1{\sqrt2}(\psi_++\psi_-),\qquad
\psi_q=\tfrac1{\sqrt2}(\psi_+-\psi_-),
\]
and
\[
G^R=-\,i\langle\psi_c\psi_q^\dagger\rangle,\qquad
G^A=-\,i\langle\psi_q\psi_c^\dagger\rangle,\qquad
G^K=-\,i\langle\psi_c\psi_c^\dagger\rangle.
\]
For the free-fermion plus Gaussian-noise model,
\[
[G^{R/A}]^{-1}(\omega)=\omega-h_{\rm eff}\pm i0^+,\qquad
\Sigma^K=2\,i\,\gamma\,J^2,
\]
and
\[
G^R(\omega)=\frac1{\omega-h_{\rm eff}+i\gamma J^2},\qquad
G^K=G^R\,\Sigma^K\,G^A.
\]
This suggests that the open-system Keldysh-Lindblad framework is not only a formal bridge between master equations and field theory, but also a direct route to sign-problem-free real-time simulation in specific dissipative lattice models [2111.04937].

Source: https://www.emergentmind.com/topics/open-system-keldysh-lindblad-framework