Non-Markovian Dynamics

Non-Markovian dynamics is open-system evolution in which the reduced system cannot be faithfully treated as evolving with no relevant memory of its past. The phrase is useful, but not unique: different communities diagnose non-Markovianity through memory kernels, failure of semigroup structure, failure of CP divisibility, information backflow, strong system-environment correlations, or structured environmental spectra.

This chapter is the navigation map for those meanings. For the article-level comparison of definitions and diagnostics, see What Non-Markovian Means. The central warning is the same throughout the chapter: do not treat “non-Markovian” as a single property that every diagnostic must agree on in every setting.

Core Idea

The exact reduced state of a system $S$ coupled to an environment $E$ is

$$\rho_S(t) = \operatorname{Tr}_E \left[ U(t,t_0)\rho_{SE}(t_0)U^\dagger(t,t_0) \right].$$

This formula is always compatible with unitary dynamics on the larger system. The hard question is whether $\rho_S(t)$ alone contains enough information to predict its own future by a memoryless reduced rule.

A time-homogeneous Markovian semigroup has

$$\Phi_0=\mathrm{id}, \qquad \Phi_{t+s}=\Phi_t\Phi_s, \qquad t,s\ge0,$$

with each $\Phi_t$ a completely positive trace-preserving map. A Lindblad–GKSL generator gives

$$\Phi_t=e^{t\mathcal L}.$$

Non-Markovian dynamics appears when this semigroup picture is too restrictive: the environment may return excitation, preserve correlations, have a long correlation time, contain a sharp resonance, or keep records that later influence the system.

Several Meanings

The word “non-Markovian” can mean different, related things.

Meaning	Diagnostic question
Memory-kernel dynamics	Does $\dot\rho(t)$ depend explicitly on earlier states $\rho(s)$?
Failure of semigroup structure	Is there no time-independent generator with $\Phi_t=e^{t\mathcal L}$?
Failure of CP divisibility	Are intermediate maps $\Phi_{t,s}$ not CPTP for some $t\ge s$?
Information backflow	Can distinguishability between two system states temporarily increase?
Strong correlations	Do system-environment correlations or bath changes remain dynamically relevant?
Structured environment	Does the bath correlation function have long tails, resonances, gaps, or recurrences?

These meanings often overlap, but they are not identical. A time-local equation with time-dependent rates can be nonsemigroup without having explicit history integrals. A memory-kernel equation can be a useful representation even when the finite-time map is completely positive. A negative instantaneous rate can signal loss of CP divisibility, but the finite-time map may still be physical on the interval of interest.

Memory-Kernel View

In a time-nonlocal description,

$$\frac{d}{dt}\rho_S(t) = \int_{t_0}^{t}ds\, \mathcal K(t,s)\rho_S(s) + \mathcal J(t).$$

The kernel $\mathcal K(t,s)$ stores how earlier reduced states influence the present derivative. The term $\mathcal J(t)$ can appear when initially discarded degrees of freedom matter.

This is the natural language for finite reservoirs, delayed feedback, structured spectra, and projection-operator derivations. The canonical pages are Memory Kernels and Nakajima–Zwanzig Projection.

Time-Local View

The same reduced map may sometimes be written as

$$\frac{d}{dt}\rho_S(t) = \mathcal K_{\mathrm{TCL}}(t)\rho_S(t),$$

even when the physics is memoryful. Invertible reduced maps admit an exact time-local generator

$$\mathcal K_{\mathrm{TCL}}(t) = \dot\Phi(t,t_0)\Phi^{-1}(t,t_0).$$

Thus “time local” is not the same as “Markovian.” The history can be compressed into time-dependent coefficients, singularities, or temporarily negative rates. See Time-Convolutionless Master Equations.

Divisibility View

For a family of maps $\Phi_{t,0}$, one asks whether intermediate maps exist:

$$\Phi_{t,0} = \Phi_{t,s}\Phi_{s,0}, \qquad t\ge s\ge0.$$

If every $\Phi_{t,s}$ can be chosen completely positive and trace preserving, the evolution is CP-divisible. In finite dimensions, a regular time-local master equation in instantaneous Lindblad form with nonnegative rates is the standard example.

If CP divisibility fails, the system’s future cannot be represented as a legitimate quantum channel acting only on the present reduced state for every interval. That is a strong and widely used notion of quantum non-Markovianity, especially in quantum information.

Information-Backflow View

For two states $\rho$ and $\sigma$, the trace distance

$$D(\rho,\sigma) = \frac12 \lVert\rho-\sigma\rVert_1$$

measures how distinguishable they are by an optimal measurement. Quantum channels contract trace distance:

$$D(\Phi(\rho),\Phi(\sigma)) \le D(\rho,\sigma).$$

If, during an open-system evolution, $D(\rho_1(t),\rho_2(t))$ temporarily increases for some pair of initial states, one interprets this as information flowing back from the environment to the system. This is a powerful intuition, but it is a diagnostic, not the only possible definition.

Physical Sources

Non-Markovian behavior is common when an environment cannot be treated as a rapidly forgetting continuum.

Source	Typical sign
finite environment	revivals and recurrences
sharp spectral feature	nonexponential decay or bound-state formation
cavity-like mode	coherent exchange before leakage
low-frequency noise	slow fluctuations and history-dependent phases
strong coupling	dressed states and failure of weak-coupling rates
correlated initial state	inhomogeneous terms and preparation dependence
correlated collision model	memory carried by repeated or correlated ancillas

These examples do not automatically invalidate every Markovian approximation. They are warnings that time-scale and correlation assumptions must be checked rather than assumed.

Relation to Existing Pages

This chapter grows out of the earlier open-system derivation pages:

• Reduced Dynamics gives the exact system-environment starting point.
• System–Bath Hamiltonians defines microscopic models.
• Born Approximation explains weak factorization.
• Markov Approximation explains the short-memory replacement and its time-scale assumptions.
• Memory Kernels treats explicit history dependence.
• Nakajima–Zwanzig Projection derives time-nonlocal equations.
• Time-Convolutionless Master Equations treats time-local generators with memory in the coefficients.
• Initial Correlations explains preparation dependence.
• Approximation Checklist lists practical validity tests.

Dedicated pages in this chapter separate What Non-Markovian Means, CP Divisibility, Information Backflow, Non-Markovianity Measures, Pitfalls in Non-Markovian Modeling, and constructive modeling methods such as Pseudomode Methods, Reaction-Coordinate Mapping, Hierarchical Equations of Motion, Collision Models, and Strong Coupling.

Canonical Boundary

This chapter owns the conceptual and finite-dimensional open-system meaning of non-Markovianity. It does not own:

• the basic definition of density operators, which belongs to Core Formalism;
• the tensor-product and partial-trace formalism, which belongs to Composite Systems and Entanglement;
• the derivation of all bath models, which is shared with later AMO, quantum matter, and many-body volumes;
• full nonequilibrium quantum field theory, influence functionals, and Schwinger–Keldysh methods, whose local bridge is Schwinger–Keldysh Bridge and whose full treatment belongs to QFT.

When a page needs a memory concept only as background, link here or to the specific canonical page rather than duplicating the diagnostic.

Common Mistakes

• Treating non-Markovianity as a single universally agreed property.
• Calling every time-dependent master equation non-Markovian.
• Calling every negative instantaneous rate unphysical.
• Assuming memory automatically makes a model more accurate.
• Forgetting that finite-time complete positivity and instantaneous generator form are different checks.
• Fitting data with a Lindblad equation and then declaring memory absent.
• Using trace-distance backflow as the only diagnostic in settings where the relevant operational task is different.
• Ignoring initial correlations and preparation dependence.

Exercises

Semigroup versus time-dependent channel

Suppose a dephasing channel has coherence factor $\eta(t)$, so $\rho_{01}(t)=\eta(t)\rho_{01}(0)$. What condition on $\eta(t)$ is required for a time-homogeneous semigroup?

Solution

The semigroup law requires

$$\eta(t+s)=\eta(t)\eta(s), \qquad t,s\ge0.$$

With continuity and $\eta(0)=1$, the usual nonzero solution is exponential:

$$\eta(t)=e^{-\Gamma t}$$

possibly with an additional Hamiltonian phase. A general nonexponential $\eta(t)$ may still define valid finite-time channels, but it is not a time-homogeneous semigroup.

Trace-distance backflow

Why does an increase of trace distance suggest information backflow?

Solution

The trace distance gives the optimal single-shot distinguishability of two quantum states. A completely positive trace-preserving map cannot increase it:

$$D(\Phi(\rho),\Phi(\sigma)) \le D(\rho,\sigma).$$

If the reduced dynamics makes $D(\rho_1(t),\rho_2(t))$ increase over some interval, the system has recovered distinguishability that was previously lost. This is naturally interpreted as information returning from environmental degrees of freedom or correlations into the accessible system.

Negative rates

In a time-local master equation with instantaneous Lindblad-like terms, a rate $\gamma(t)$ becomes negative over a short interval. What can and cannot be concluded immediately?

Solution

A negative instantaneous rate means the generator is not in ordinary Markovian Lindblad form during that interval and, under the usual regularity assumptions, the evolution is not CP-divisible there. It does not immediately prove that the finite-time map from the initial time to $t$ is unphysical. The finite-time map must be checked directly, for example through a Choi matrix or an exact dilation.

Cross-Links

• Applications and Experimental Platforms
• Computational Notebooks
• Reference
• Reduced Dynamics
• What Non-Markovian Means
• CP Divisibility
• Information Backflow
• Non-Markovianity Measures
• Pseudomode Methods
• Reaction-Coordinate Mapping
• Hierarchical Equations of Motion
• Collision Models
• Strong Coupling
• Pitfalls in Non-Markovian Modeling
• Markov Approximation
• Memory Kernels
• Nakajima–Zwanzig Projection
• Time-Convolutionless Master Equations
• Initial Correlations
• Lindblad–GKSL Equation
• Quantum Dynamical Semigroups
• Completely Positive Maps
• Schwinger–Keldysh Bridge
• Approximation Checklist

References

• H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford University Press (2002).
• H.-P. Breuer, E.-M. Laine, J. Piilo, and B. Vacchini, “Colloquium: Non-Markovian dynamics in open quantum systems,” Reviews of Modern Physics 88, 021002 (2016).
• Á. Rivas, S. F. Huelga, and M. B. Plenio, “Quantum non-Markovianity: characterization, quantification and detection,” Reports on Progress in Physics 77, 094001 (2014).
• I. de Vega and D. Alonso, “Dynamics of non-Markovian open quantum systems,” Reviews of Modern Physics 89, 015001 (2017).
• D. Chruściński, A. Kossakowski, and Á. Rivas, “Measures of non-Markovianity: Divisibility versus backflow of information,” Physical Review A 83, 052128 (2011).
• S. Nakajima, “On quantum theory of transport phenomena,” Progress of Theoretical Physics 20, 948–959 (1958).
• R. Zwanzig, “Ensemble method in the theory of irreversibility,” Journal of Chemical Physics 33, 1338–1341 (1960).