Generalized Born-Markov Approximation

• Generalized Born–Markov approximation is a family of extended treatments for open quantum systems that retain finite memory, off-diagonal couplings, and dispersive corrections typically lost in standard methods.
• It derives non-secular Lindblad and Redfield equations, time-local expansions, and self-consistent master equations to capture transient coherences and renormalized decay rates.
• Practical implementations span quantum transport, driven systems, and numerically exact schemes, offering improved accuracy over conventional weak-coupling and memoryless approximations.

Searching arXiv for the cited works and closely related formulations of generalized Born–Markov treatments. Generalized Born–Markov approximation denotes a family of extensions and reinterpretations of the conventional Born–Markov treatment of open quantum systems. In the literature considered here, the phrase is used for several closely related constructions: non-secular Lindblad-like and Redfield equations with fully complex bath-induced coefficients (Liao et al., 2019), time-local hierarchies that interpolate between Born and Markov limits (Karlewski et al., 2014), self-consistent Born master equations for transport (Liu et al., 2013), exact semigroup constructions with exact quantum regression for specially engineered reservoirs (Lonigro et al., 2022), and enlarged-system or finite-memory formulations that go beyond both the Born–Markov and secular approximations (Uchiyama, 2023). Across these usages, the common theme is the systematic retention of structures that the standard weak-coupling, memoryless, secular approximation discards: finite-time memory, off-diagonal Bohr-frequency couplings, dispersive Lamb-shift contributions, self-consistent level broadening, and higher-order system–bath correlations.

1. Standard Born–Markov structure as the reference point

For a total Hamiltonian

$$H=H_s+H_b+H_I,\qquad H_I=\sum_\alpha S_\alpha\otimes E_\alpha,$$

the Born–Markov master equation obtained under weak coupling, a factorized initial state $\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$, and a bath correlation time short compared to the system evolution takes the Schrödinger-picture form

$$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$

with

$$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$

This equation is the starting point for the non-secular Lindblad and Redfield forms developed by Liao and Liang (Liao et al., 2019).

In transport problems, the same reference structure appears after tracing over reservoirs and expanding to second order in the tunneling Hamiltonian. If reservoir correlations are replaced by delta functions and $\rho(\tau)\to\rho(t)$, the self-consistent Born formulation reduces back to the usual second-order Born–Markov master equation in Lindblad or Redfield form,

$$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$

valid under weak coupling, large bias, wide-band reservoirs, and high temperature (Liu et al., 2013).

The standard secular approximation is an additional step beyond Born–Markov. It removes couplings between distinct Bohr-frequency sectors and typically yields a time-independent GKSL generator. Generalized Born–Markov constructions are distinguished precisely by relaxing some part of that simplification.

2. Non-secular Lindblad and Redfield forms with generalized coefficients

A central usage of the term is the derivation of non-secular Lindblad-like and Redfield equations directly from the Born–Markov master equation without invoking the usual secular approximation (Liao et al., 2019). Introducing the system eigenoperators

$$F_\alpha(\omega)=\sum_{\epsilon_n-\epsilon_m=\omega}\Pi(\epsilon_m)S_\alpha\Pi(\epsilon_n),$$

so that

$$S_\alpha=\sum_\omega F_\alpha(\omega), \qquad S_\alpha(\tau)=\sum_\omega e^{-i\omega\tau}F_\alpha(\omega),$$

and the spectral correlation tensor

$$\Gamma_{\alpha\beta}(\omega) \equiv \int_0^\infty d\tau\, e^{i\omega\tau}W_{\alpha\beta}(\tau) = \tfrac12\gamma_{\alpha\beta}(\omega)+i\,T_{\alpha\beta}(\omega),$$

one obtains a non-secular Lindblad-like equation

$$\frac{d\rho_s}{dt} = -i[H_s+H_{ls},\rho_s]+D(\rho_s),$$

with

$\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$0

$\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$1

and fully complex coefficients

$\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$2

The real and imaginary parts decompose as

$\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$3

$\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$4

In the corresponding non-secular Redfield form, the Redfield tensor is built from the same $\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$5 without discarding its imaginary part.

This construction differs sharply from the traditional real-coefficient and secular treatment. The secular approximation sets $\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$6 whenever $\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$7 or $\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$8 appears and drops all imaginary parts of $\rho_{\mathrm{tot}}(0)=\rho_s(0)\otimes\rho_b$9, so that $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$0 and $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$1. Consequently, the traditional Lindblad and Redfield equations lose off-diagonal relaxation channels with $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$2 and remove dispersive corrections except for a trivial shift. Physically, the retained non-secular complex terms account for transient coherence between different transition frequencies and for small frequency shifts induced by the bath; these effects can be crucial whenever Bohr frequencies are nearly degenerate or the bath spectrum is structured (Liao et al., 2019).

The dynamical consequences depend on the model. For a weakly coupled three-level system with well-separated levels, non-secular and secular treatments yield only minor differences in populations and coherences, and ignoring $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$3 distorts decay rates only slightly. For the PE545 photosynthetic complex, however, the secular approximation lengthens the relaxation time by nearly one order of magnitude, and omitting $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$4 compresses the dynamics time-scale by up to two orders of magnitude; the full non-secular complex-coefficient master equations recover the true Born–Markov evolution, including transient beatings and correct thermalization times (Liao et al., 2019).

3. Time-local expansions and the coupling–memory hierarchy

A second major meaning of generalized Born–Markov approximation is the systematic construction of time-local master equations that organize corrections simultaneously in the coupling strength and in the degree of memory retention. Karlewski and Marthaler derive an exact time-local expansion

$$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$5

where each $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$6 collects all contributions of order $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$7 (Karlewski et al., 2014). The first terms read

$$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$8

Here $$\frac{d\rho_s}{dt} = -i[H_s,\rho_s] + \sum_{\alpha\beta}\int_0^\infty d\tau\, W_{\alpha\beta}(\tau) \left\{ S_\beta(-\tau)\rho_s S_\alpha - S_\alpha S_\beta(-\tau)\rho_s \right\} +\mathrm{H.c.},$$9 is a fourth-order Born memory term, while $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$0 is the leading non-Markovian correction. Their power counting is identical:

$$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$1

The principal conclusion is that non-Markovian terms are of the same order of magnitude as higher-order terms in the system–bath coupling. This is why, in that framework, analyzing non-Markovian behaviour implies going beyond the Born approximation (Karlewski et al., 2014).

The same formalism also resolves the initial-state problem of non-Markovian master equations. When the exact memory integral is split into contributions from $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$2 and $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$3, the initial-correlation pieces cancel order by order in the full expansion. If the negative-time branch is omitted, the missing terms appear as short-lived initial-slip corrections decaying on the bath-correlation timescale.

Quantum Brownian motion provides a complementary formulation of the same issue. In the Drude–Ohmic model, a systematic derivative expansion of the nonlocal Born master equation yields a hierarchy of local-in-time equations with $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$4 inertial corrections, but comparison with the exact Heisenberg–Langevin solution shows that the Born approximation is reliable only if

$$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$5

The second condition is more stringent than the usual weak-coupling requirement: it demands that the frequency renormalization $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$6 remain small compared with $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$7. In that analysis, the steady-state interaction energy $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$8 serves as a proxy for the system–bath correlations missed by Born factorization, and its dependence on $$W_{\alpha\beta}(\tau)\equiv \mathrm{Tr}_b\{E_\alpha(\tau)E_\beta\rho_b\}, \qquad S_\alpha(\tau)=e^{iH_s\tau}S_\alpha e^{-iH_s\tau}.$$9 tracks the discrepancy between exact and Born-based results (Boyanovsky et al., 2017).

4. Alternative generalized constructions in transport, regression, and driven systems

Beyond non-secular complex-coefficient master equations, the term also labels several other constructions that modify the Born–Markov strategy in distinct ways.

Construction	Key move	Characteristic consequence
n-SCBA-ME (Liu et al., 2013)	Replace the free propagator inside the memory integral by a dressed propagator $\rho(\tau)\to\rho(t)$0	Non-Markovian kernel, level broadening, cotunneling, many-body correlations, and recovery of the nonequilibrium Kondo effect
Extended global TCL2 (Uchiyama, 2023)	Keep finite-time memory integrals $\rho(\tau)\to\rho(t)$1 and retain $\rho(\tau)\to\rho(t)$2 terms	Short-time positivity, oscillatory features hidden by SA, and a stationary state very near the Gibbs state of $\rho(\tau)\to\rho(t)$3
Generalized spin-boson semigroup (Lonigro et al., 2022)	Choose flat, unbounded spectra $\rho(\tau)\to\rho(t)$4 on $\rho(\tau)\to\rho(t)$5	Exact semigroup $\rho(\tau)\to\rho(t)$6 and exact quantum-regression hierarchy
CC–Floquet–FCS embedding (Restrepo et al., 2017)	Absorb a collective reservoir coordinate into an enlarged supersystem and then apply Born–Markov to the residual baths	Access to structured spectra, non-secular Sambe-space dynamics, and four operating regimes of a driven thermal machine

In quantum transport, the self-consistent Born approximation replaces the free propagator $\rho(\tau)\to\rho(t)$7 by a dressed propagator that already contains second-order self-energy corrections. The resulting master equation is essentially non-Markovian and incorporates the interplay of multi-tunneling processes and many-body correlations. For steady state it recovers the exact result of noninteracting transport under arbitrary voltages and the nonequilibrium Kondo effect, while retaining computational efficiency for shot noise (Liu et al., 2013).

In the generalized spin-boson setting, a flat spectrum on the full real line produces an exact time-homogeneous semigroup without any Born or Markov limiting procedure. In that single case, the full unitary dynamics factorizes in such a way that the quantum-regression hierarchy is exactly satisfied at all orders. No other choice of spectral density on $\rho(\tau)\to\rho(t)$8 yields an exact semigroup plus exact regression at all times, apart from trivial re-parametrizations (Lonigro et al., 2022). This usage is conceptually distinctive: “generalized Born–Markov” refers not to a perturbative approximation but to an exactly solvable memoryless model.

In periodically driven thermal machines, a collective-coordinate mapping embeds strong-coupling and non-Markovian features of a structured bath into an enlarged supersystem, after which Floquet theory and full counting statistics yield a time-independent Liouvillian in extended space. Scanning the detuning under the red-sideband resonance reveals four operating regimes: heat engine, refrigerator, heating-of-cold only, and dissipator-regime. As the coupling strength with one bath is increased, the refrigerator regime disappears, the heat engine regime narrows, and their efficiency and coefficient of performance decrease (Restrepo et al., 2017).

5. Beyond analytic master equations: finite-memory and numerically exact schemes

Several works position generalized Born–Markov methods relative to explicitly beyond-Born–Markov approaches. One numerically exact strategy discretizes the bath into $\rho(\tau)\to\rho(t)$9 modes, truncates the bath basis by a phonon cutoff $$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$0, and propagates the full system–bath state with a short-iterative Lanczos scheme combined with exact diagonalization (Cangemi et al., 2018). In that construction, the initial state remains factorized, but neither the Born approximation nor the Markov approximation is invoked. The only approximations are bath discretization and truncation of mode occupations; by increasing $$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$1 and $$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$2, the exact continuum result is recovered in principle. This framework includes memory effects, multi-phonon and vertex processes to arbitrarily high order up to the chosen cutoff, and yields the reduced Gibbs state

$$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$3

whereas the Born–Markov Lindblad equation relaxes to the factorized Gibbs state of the bare qubit (Cangemi et al., 2018).

A recent qubit-reset study uses the phrase in yet another sense: as a sketch of a finite-memory, polaron-corrected master equation motivated by numerically exact tensor-network and time-dependent variational-principle simulations (Ortega-Taberner et al., 10 Mar 2026). There the proposed generalized Born–Markov equation is written in a polaron frame as

$$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$4

with renormalized rates

$$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$5

Its stated role is to capture two effects seen in the exact simulations: reduced effective dissipation from polaron dressing and non-exponential relaxation from finite memory. This suggests an emerging trend in which “generalized Born–Markov” increasingly denotes reduced models that retain explicit correlation-induced renormalization while remaining much cheaper than fully non-Markovian simulations.

6. Validity, positivity, and recurrent misconceptions

The domain of validity depends on which generalized construction is meant. In the non-secular Lindblad and Redfield derivation from the Born–Markov equation, one still requires weak system–bath coupling and a bath correlation time short compared to the system evolution time. Within that regime, the non-secular complex-coefficient approach is strongly recommended whenever Bohr frequencies are nearly degenerate or the bath has sharp spectral features, while for well-separated levels and broad, smooth spectral densities the secular real-coefficient Lindblad equation is often adequate (Liao et al., 2019).

Positivity is a recurring issue. In the generalized global TCL2 approach, the second-order kernels behave at very small $$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$6 as

$$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$7

and because $$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$8 is a positive-semidefinite matrix, the instantaneous dissipator is of GKSL form with strictly positive rates proportional to $$\dot\rho(t) = -\,i[H_S,\rho(t)] +\sum_{\alpha\mu\nu} \Gamma_{\alpha\mu\nu} \left( a_\nu\rho a_\mu^\dagger -\tfrac12\{a_\mu^\dagger a_\nu,\rho\} \right) +\mathrm{H.c.},$$9. This is why the map remains completely positive for times of order the bath correlation time. By contrast, the time-independent Redfield generator obtained after taking $$F_\alpha(\omega)=\sum_{\epsilon_n-\epsilon_m=\omega}\Pi(\epsilon_m)S_\alpha\Pi(\epsilon_n),$$0 while retaining non-secular terms may violate positivity at short simulation times (Uchiyama, 2023). In the complex-coefficient non-secular Lindblad construction, positivity is preserved under the same conditions as the secular Lindblad form provided the full complex coefficients are used; dropping $$F_\alpha(\omega)=\sum_{\epsilon_n-\epsilon_m=\omega}\Pi(\epsilon_m)S_\alpha\Pi(\epsilon_n),$$1 or mixing secular with partial imaginary parts can violate complete positivity (Liao et al., 2019).

A persistent misconception is that CP-divisibility and quantum regression are equivalent indicators of Markovianity. They are not. For generalized spin-boson amplitude-damping channels, CP-divisibility means that a family of CPTP maps admits CPTP propagators $$F_\alpha(\omega)=\sum_{\epsilon_n-\epsilon_m=\omega}\Pi(\epsilon_m)S_\alpha\Pi(\epsilon_n),$$2, and in the TCL form it is equivalent to nonnegative rates $$F_\alpha(\omega)=\sum_{\epsilon_n-\epsilon_m=\omega}\Pi(\epsilon_m)S_\alpha\Pi(\epsilon_n),$$3. Quantum regression is strictly stronger: it requires the exact hierarchy of multi-time correlation functions to factor through the single-time reduced map, which generally fails once system–bath correlations develop. Exact regression holds in the flat-spectrum semigroup model precisely because the microscopic dynamics is engineered so that this stronger property is satisfied (Lonigro et al., 2022).

Another misconception is methodological: non-Markovian corrections can be appended to a second-order Born treatment without simultaneously addressing higher-order coupling effects. The time-local expansion of Karlewski and Marthaler shows the opposite. The leading non-Markovian term and the leading fourth-order Born term are parametrically of the same order, so a consistent analysis of non-Markovian behaviour requires going beyond Born approximation as well (Karlewski et al., 2014).

Taken together, these results show that “generalized Born–Markov approximation” is not a single formula but a technical program. Depending on context, it may mean retaining non-secular complex coefficients, organizing memory and coupling corrections in a common hierarchy, dressing propagators self-consistently, embedding structured environments into enlarged Markovian supersystems, or identifying exactly solvable models whose dynamics is memoryless without any limiting procedure.