An open quantum system is a physical system whose interaction with surrounding degrees of freedom cannot be neglected in describing its behavior. Examples include an excited atom coupled to electromagnetic radiation, an electron spin interacting with a solid, and a superconducting circuit connected to control and measurement equipment. Unlike an ideal isolated system, its state generally does not evolve according to a Schrödinger equation involving only the system’s own Hamiltonian. Instead, environmental interactions can produce relaxation, fluctuations, loss of coherence, and correlations with inaccessible degrees of freedom.
The theory of open quantum systems combines quantum mechanics, statistical mechanics, and information-theoretic descriptions of physical processes. It explains how irreversible behavior can emerge from reversible microscopic dynamics, establishes practical models for spectroscopy and transport, and supplies the mathematical language for noise and measurement in quantum technologies. Its central objects are reduced density operators, quantum channels, and master equations; their validity depends on assumptions about environmental correlations, coupling strength, and relevant observation times.
Subsystems, environments, and reduced states
The distinction between system and environment is a modeling choice rather than an intrinsic division of nature. In an optical experiment, the system might consist of one atom, a selected pair of atomic levels, or an atom together with a cavity mode. Everything excluded from that description belongs to the environment. Enlarging the system can turn an apparently memory-dependent interaction into a simpler dynamical model, although the enlarged calculation becomes more expensive. An environment need not be large, thermal, or uncontrollable: another qubit can serve as an environment for a selected subsystem.
For an ordinary tensor-product decomposition, the joint state acts on . The total Hamiltonian operator is conventionally written
where describes the interaction. If this combined system is isolated, its state evolves unitarily,
The state available to observations on the system alone is its reduced density operator,
The partial trace removes environmental indices while preserving every expectation value of a system observable:
A reduced state may be mixed even when the joint state is pure. For example, the entangled two-qubit state
gives . The subsystem’s mixedness then reflects quantum entanglement, not merely uncertainty about which pure state was prepared. Local measurements cannot distinguish this reduced state from an appropriately prepared classical mixture; joint measurements can distinguish the underlying situations.
Tracing out an environment does not itself change the joint physical state. Rather, it changes which information the description retains. Interactions can transfer information about a subsystem into system–environment correlations, making it unavailable to local measurements. This accounts for both the usefulness and the limitations of an effectively irreversible description. A finite environment may later return information and produce revivals, whereas a large environment with many incommensurate frequencies can suppress observable recurrences over practical times.
The environment also influences which system properties remain stable. An interaction that distinguishes position states tends to suppress spatial coherence; one that distinguishes energy levels can suppress coherence between those levels. Consequently, openness is not synonymous with heating or particle loss. A subsystem may retain constant mean energy while its interference properties change substantially.
Quantum channels and physically admissible evolution
A quantum channel describes a deterministic transformation of a quantum state without specifying every microscopic environmental detail. For a fixed initial environmental state , an initially uncorrelated preparation gives
The map is linear, trace preserving, and completely positive. These conditions ensure that density operators remain valid states, including when the system is entangled with an otherwise unaffected auxiliary system.
Positivity alone means that positive operators are mapped to positive operators. Complete positivity requires the extended map to remain positive for an auxiliary space of any dimension. This stronger condition matters because a locally applied operation must remain physically consistent on an entangled state. Matrix transposition illustrates the distinction: it preserves positivity on a single system but is not completely positive, so it cannot represent a universally implementable deterministic quantum operation.
In finite dimensions, every completely positive, trace-preserving map has an operator-sum representation,
The are Kraus operators. Their representation is not unique: different collections can describe the same channel. A channel also admits a realization through a unitary interaction with an auxiliary system followed by discarding that auxiliary system. This dilation connects the operational description of noise to the microscopic description of an environment.
For a two-level system, amplitude damping can be represented by
Here is the probability that an initially excited state decays during the interval represented by the channel. The excited population is multiplied by , while the off-diagonal coherence is multiplied by . A dephasing channel instead reduces coherence without changing populations in its preferred basis.
The product-state assumption has an important boundary. If system and environment begin correlated, specifying alone generally does not determine : different joint states with the same marginal can yield different futures. A reduced map defined only on a restricted family of preparations need not extend to a completely positive map on every possible system state. This does not invalidate complete positivity for freely implementable operations; it shows that the preparation procedure and its environmental correlations are part of the physical specification.
Markov approximations and Lindblad master equations
A quantum master equation specifies the time evolution of a reduced density operator. The most widely used memoryless form is the Gorini–Kossakowski–Sudarshan–Lindblad equation,
The anticommutator is , the rates satisfy , and may contain environmental energy shifts. The are jump or noise operators. With a time-independent generator, this equation defines a completely positive, trace-preserving semigroup:
In finite dimensions, the generator of every norm-continuous quantum dynamical semigroup has this form.
A common microscopic derivation begins with
Environmental correlation functions,
determine how the bath affects system transitions. Their Fourier transforms supply transition rates at the system’s Bohr frequencies. These frequencies are energy differences divided by , where is the reduced Planck constant. The bath therefore matters not simply through its overall strength, but through the fluctuations it possesses at frequencies to which the system responds.
Three approximations often enter this derivation, but they serve different purposes. The Born approximation treats weak coupling perturbatively and neglects certain changes in the environmental state. The Markov approximation assumes that environmental correlations decay much faster than the reduced state changes, permitting a memory integral to be replaced by a time-local expression. A secular approximation then averages rapidly oscillating terms associated with sufficiently separated transition frequencies. This last step often produces a manifestly completely positive generator.
These approximations require compatible time scales. A coarse-graining interval must be long enough to average environmental correlations yet short compared with dissipative relaxation. Full secularization additionally requires relevant frequency differences to be resolvable on that interval. Nearly degenerate transitions can violate the latter condition even when coupling is weak. Conversely, a rapidly relaxing environment can support a useful Markov description without being at high temperature.
A second-order equation without secularization is commonly called a Redfield equation. It can preserve interference between nearby transitions more accurately in some regimes, but it is not generally completely positive and can produce unphysical states outside its domain of validity. The distinction is therefore not simply between an accurate non-Lindblad equation and an inaccurate Lindblad equation. Accuracy depends on time scales and preparation, while complete positivity supplies a separate consistency requirement for a model intended to act on arbitrary states.
Relaxation, dephasing, and elementary models
Two fundamental environmental effects are energy relaxation and decoherence. Relaxation changes populations and typically transfers energy between system and environment. Dephasing suppresses off-diagonal density-matrix elements in a specified basis. Decoherence is the broader loss of observable phase relationships, including contributions from both relaxation and pure dephasing. These distinctions are operational: a reduced interference contrast does not by itself identify which physical process caused it.
For a two-level system with
the Pauli matrices provide a compact description. Coupling to a thermal electromagnetic environment can be modeled by
where
The rates and describe downward and upward transitions, while is the pure-dephasing rate under this convention. For a bath at inverse temperature , thermal detailed balance gives
The population-relaxation and coherence-decay times are then
Thus within this exponential Markovian model. This relation is central to nuclear magnetic resonance and qubit characterization. It should not be applied indiscriminately to nonexponential decay or to , which commonly includes reversible inhomogeneous broadening. Echo sequences can refocus some slowly varying frequency offsets but cannot generally undo energy lost into an uncontrolled environment.
An excited atom emits a photon even when the surrounding radiation field begins in its vacuum state. In a weak-coupling description, the electromagnetic vacuum supplies a continuum of accessible final states, and the excited population approximately decays exponentially. The same interaction also shifts the transition frequency. Within the rotating-wave description of vacuum amplitude damping, an atom initially in its ground state remains there: vacuum fluctuations do not constitute a supply of ordinary thermal excitation energy.
Another standard example is a damped quantum harmonic oscillator,
Here , and the mean occupation obeys
This model describes thermalization of optical, microwave, and mechanical modes, while also showing explicitly how dissipative drift and environmental fluctuations appear together.
Memory effects and non-Markovian dynamics
Environmental memory becomes important when information stored outside the system influences its later evolution. Examples include an atom coupled to a narrow cavity resonance, a spin interacting with a small spin environment, and a quantum emitter near a photonic band edge. Strong coupling, slowly decaying environmental correlations, and repeated encounters with the same environmental degrees of freedom can all invalidate a simple memoryless approximation. Quantum non-Markovianity consequently covers several related but inequivalent notions of memory.
One exact organizational form is the Nakajima–Zwanzig equation,
where the memory kernel determines how earlier states contribute to the present derivative. The term can encode initial correlations for the chosen projection. In practice, exact kernels are rarely easy to calculate, but the equation identifies what a Markov approximation discards.
Memory-dependent evolution can also admit a time-local representation,
when an appropriate dynamical map is invertible. Time locality therefore does not by itself imply absence of memory. A time-dependent generator can contain temporarily negative canonical decay rates while the overall map from the initial time remains completely positive. Singularities of a time-local generator may occur when the dynamical map ceases to be invertible, even though the underlying physical evolution remains regular.
An important criterion is complete-positive divisibility. A family is CP-divisible if, for every , there is a completely positive, trace-preserving map such that
Another criterion examines the distinguishability of two initial states through their trace distance,
An increase in this distance under the same reduced dynamics is interpreted as information backflow. However, trace-distance backflow and failure of CP-divisibility are not equivalent in full generality; auxiliary entangled systems can reveal distinctions unavailable to system-only tests.
Multitime experiments introduce a further issue. A family of one-time state maps does not determine how a system responds to a sequence of interventions. A process tensor encodes the dependence of later outcomes on earlier preparations, operations, and measurements, including correlations transmitted through the environment. This makes memory an experimentally testable property of a sequence of interactions rather than solely a feature inferred from a fitted decay curve. Observed oscillations or revivals can provide clues, but coherent driving and unresolved degrees of freedom must be distinguished from genuine environmental memory.
Continuous measurement and quantum trajectories
An unconditional master equation describes the state when environmental measurement records are ignored. If outgoing photons, electrical signals, or other environmental observables are monitored, the relevant state is instead conditioned on the observed record. A quantum trajectory is one realization of this conditional stochastic evolution. Averaging over all possible records recovers the unconditional density operator, provided the monitoring scheme and detector inefficiencies are modeled consistently.
For photon counting, a Lindblad term with operator can be unraveled into periods of conditional evolution interrupted by jumps. During a short interval , a count occurs with probability
and the conditional state changes to
Between counts, the unnormalized state evolves using an effective non-Hermitian Hamiltonian,
Its decreasing norm gives the probability that no count has occurred. Renormalization then produces the conditional no-count state.
Absence of a detected event can itself carry information. If an excited atom would probably emit during a given interval, a sufficiently long interval without a count changes the relative probabilities assigned to its states. The resulting state update does not require an instantaneous energy-emission event. It reflects conditioning on an informative measurement outcome. With imperfect detection, unobserved emissions must also be included, and the conditioned state need not remain pure.
Other detection arrangements produce different trajectories for the same unconditional master equation. Homodyne detection mixes the outgoing field with a strong local oscillator and measures a field quadrature. Its record is approximately continuous and noisy, leading to a diffusive stochastic master equation rather than discrete jumps. Thus a particular trajectory is not a unique hidden history assigned by the master equation alone; it depends on what environmental information is measured and how that information is processed.
Trajectory methods connect open-system dynamics to quantum filtering, state estimation, and feedback. Measurement-based feedback changes a control field in response to the observed record, whereas coherent feedback routes quantum signals without first reducing them to classical data. These descriptions are also related to the measurement problem: they specify outcome probabilities, conditioning, and backaction, but do not by themselves settle every interpretive question about individual outcomes. Computationally, trajectories provide a separate advantage because evolving many state vectors can be cheaper than evolving one full density matrix, especially when only a few observables are required.
Thermalization and quantum thermodynamics
Open-system theory gives a dynamical account of contact with thermal reservoirs. For a weakly coupled system in thermal equilibrium at temperature , the expected stationary state is
Here is the partition function. Obtaining this state requires appropriate relations between upward and downward transition rates. A generic Lindblad equation is not necessarily thermal: it may describe dephasing, particle loss, external pumping, or an engineered reservoir with no single equilibrium temperature.
A thermal stationary state is also not automatically the unique long-time state. Conserved quantities, disconnected transition sectors, and dark states can prevent complete relaxation. If the dynamics preserves several independent subspaces, the final state may retain information about the initial population in each sector. Statements that a bath “thermalizes a system” therefore require conditions on both its transition rates and the connectivity of the dynamics.
In a common weak-coupling convention, the internal energy is
Its derivative separates as
The first term is associated with energy exchange caused by state changes, and the second with work supplied by external control. For a thermal dissipator, its contribution to the first term is conventionally called a heat current. With several baths, each dissipator can be assigned a separate current when the microscopic approximations support that decomposition.
For a thermal Markovian generator satisfying suitable equilibrium conditions, the second law of thermodynamics can be expressed as
This entropy-production inequality is related to the decrease of quantum relative entropy toward the stationary thermal state. It constrains quantum engines, refrigerators, and transport devices. The entropy of the system alone need not increase: cooling can reduce it while the reservoir’s entropy increases by a greater amount.
Strong coupling complicates these assignments because interaction energy cannot be neglected. The reduced equilibrium state can differ from and can instead be characterized through a Hamiltonian of mean force. Likewise, deriving separate “local” dissipators for interacting subsystems may produce incorrect heat currents if internal coupling is important. Thermodynamic consistency must therefore be assessed alongside state positivity and numerical accuracy, particularly when reservoir temperatures differ or controls vary rapidly.
Experimental platforms and engineered dissipation
Open-system methods connect experimentally measurable decay and noise to microscopic interactions. In spectroscopy, environmental effects appear as linewidths, frequency shifts, and changes in response functions. Under appropriate stationarity and Markov assumptions, the quantum regression theorem evolves certain two-time correlation functions using the same generator as single-time expectation values. Their Fourier transforms determine fluorescence spectra and other signals. Applying this theorem outside those assumptions can fail even when a master equation fits one-time populations well.
In cavity quantum electrodynamics, an atom exchanges excitation with a confined electromagnetic mode while both couple to additional degrees of freedom. When coherent coupling dominates relevant decay rates, excitation can oscillate between atom and cavity. In another regime, a rapidly decaying cavity can be eliminated to produce an effective atomic decay channel. Whether the cavity belongs to the system or the environment therefore depends on both the physical parameters and the observables of interest.
Superconducting circuits, trapped ions, semiconductor spins, and neutral atoms provide distinct realizations of environmental coupling. In a solid, phonons can carry energy away from an electronic excitation; fluctuating magnetic or electric fields can modulate a transition frequency. In a trapped-ion experiment, motional modes may be controlled resources, effective reservoirs, or unwanted noise sources. Noise characterization uses relaxation measurements, echo experiments, and controlled modulation to distinguish spectral components and identify coupling mechanisms.
For quantum computing, environmental interaction is a source of errors but not merely a single decay constant. Correlated errors, leakage outside a computational subspace, and slow parameter drift can matter differently from independent amplitude damping or dephasing. Quantum error correction protects encoded information by extracting error syndromes while avoiding direct measurement of the logical state. Its performance depends on the structure and time dependence of the actual noise, not only on an average gate-error probability.
Dissipation can also be engineered as a resource. Designed jump operators may drive a system toward a target entangled state or a protected manifold. A pure state annihilated by all relevant jump operators is dark to those channels, although Hamiltonian evolution must also preserve the target for it to remain stationary. In driven many-body systems, competition between coherent interactions and loss can generate collective steady states, bistability, or dissipative phase transitions. These transitions concern steady-state properties in an appropriate large-system limit; long-lived switching in a finite apparatus does not alone establish a sharp phase transition.
Historical development, computation, and model validation
Open-system theory developed from several lines of research rather than a single founding model. Einstein’s 1917 treatment of absorption and emission coefficients linked radiation-induced transitions to thermal equilibrium. Later work on magnetic resonance and irreversible statistical mechanics established relaxation equations and systematic perturbative techniques. Sadao Nakajima in 1958 and Robert Zwanzig in 1960 developed projection-operator approaches that separated selected variables from unresolved degrees of freedom. Richard Feynman and Frank Vernon introduced the influence-functional approach in 1963, expressing environmental effects within a path-integral description.
A decisive mathematical result appeared in 1976, when Vittorio Gorini, Andrzej Kossakowski, and E. C. George Sudarshan, and independently Göran Lindblad, characterized generators of completely positive quantum dynamical semigroups. This connected phenomenological damping equations to precise requirements on admissible state evolution. Weak-coupling limit methods, notably associated with E. B. Davies, supplied a rigorous connection between microscopic Hamiltonians and Markovian generators under suitable scaling assumptions. Later developments in quantum optics and information theory made channels, monitoring, and entanglement central organizing concepts.
For computation, a -dimensional Hilbert space requires density-matrix entries. Directly representing the corresponding superoperator can require order entries, although sparse structure often reduces storage. Many-body Hilbert spaces themselves grow exponentially with particle number, so density-operator methods become costly quickly. Quantum trajectories reduce the state-storage burden but introduce sampling error; tensor-network methods exploit restricted correlation or entanglement structure, whose usefulness varies with the dynamics.
Memory-dependent calculations use several complementary strategies. Hierarchical equations of motion introduce auxiliary operators associated with structured environmental correlations and can converge systematically when their decomposition and hierarchy are adequately resolved. Influence-functional and tensor-network methods retain time correlations explicitly. Reaction-coordinate mappings move a strongly coupled collective environmental mode into an enlarged system, while stochastic and perturbative methods organize different approximations to residual fluctuations. No single approach is efficient for every spectral density, temperature, coupling strength, and simulation time.
Validation therefore requires separate physical and numerical checks. Trace, Hermiticity, and positivity test whether computed states remain admissible, but do not establish that the assumed environment is correct. Numerical convergence requires varying time steps, basis cutoffs, hierarchy depth, or trajectory number. Physical validation can compare decay curves, spectra, stationary populations, multitime correlations, and responses to intervention. A model that reproduces one exponential relaxation curve may still misidentify thermal populations or environmental memory; independent observables are needed to distinguish mechanisms that produce similar single-time behavior.