Baths, Reservoirs, and Environments

An open-system model begins by choosing which degrees of freedom are the system and which are not. The words environment, bath, reservoir, noise source, and record describe different roles for the degrees of freedom outside the system. They are often used loosely, but they do not carry identical assumptions.

The terms overlap and usage varies between fields. They describe the approximation and the question, rather than disjoint kinds of physical objects.

Helpful background. Reduced dynamics explains what is lost by ignoring an environment; system–bath Hamiltonians specify the coupling.

Environment

An environment is simply whatever is not included in the chosen system. If

$$\mathcal H_{\mathrm{tot}} = \mathcal H_S\otimes\mathcal H_E,$$

then $E$ is the environment relative to the system $S$.

That definition is deliberately broad. An environment can be:

• a thermal electromagnetic field;
• phonons in a solid;
• gas molecules scattering from a particle;
• a finite spin register;
• a detector pointer;
• unobserved ancillas in a circuit;
• a classical drive whose fluctuations are modeled statistically;
• another subsystem that will later be measured.

The word environment alone says nothing about equilibrium, Markovianity, temperature, size, Gaussianity, or irreversibility. It only states a modeling boundary.

This boundary is not unique. A strongly coupled cavity mode, vibrational coordinate, reaction coordinate, or detector memory may be treated as part of the environment in one model and part of the enlarged system in another.

Bath

In common weak-coupling models, a bath is an environment described by a reference state and assumed to be only weakly perturbed in the properties relevant to the system. The reference state is often denoted $\rho_B$ or $\eta_E$ and is usually assumed to be stationary:

$$[H_B,\rho_B]=0.$$

Stationarity implies that bath correlations depend only on time differences. For a bath operator $B(t)$,

$$C_{BB}(t-s) = \operatorname{Tr}_B \left[ B(t)B(s)\rho_B \right].$$

A bath model typically assumes that the system does not appreciably change the bath’s macroscopic state over the timescale of interest. Together with a controlled interaction expansion, this can support a weak-coupling master equation; it does not make all system–bath correlations vanish. In a common starting point,

$$\rho_{SB}(0) = \rho_S(0)\otimes\rho_B,$$

and perturbation theory tracks how the bath correlations affect $\rho_S(t)$.

The word bath does not imply thermal or Markovian. Squeezed and driven bath states may be stationary only in a suitable rotating frame, or genuinely time dependent. A stationary structured bath can retain memory for a long time.

Reservoir

A reservoir is a bath used as a source or sink for a conserved quantity while its intensive parameters remain approximately fixed. The conserved quantity might be energy, particles, angular momentum, charge, or excitations in a mode family.

The standard thermodynamic examples are:

• a heat reservoir with fixed temperature $T$;
• a particle reservoir with fixed chemical potential $\mu$;
• an electronic lead with fixed temperature and voltage;
• an optical or microwave transmission line that supplies or absorbs photons;
• an engineered dissipative reservoir used to stabilize a target state.

For a thermal reservoir,

$$\rho_B = \frac{e^{-\beta H_B}}{Z_B}, \qquad \beta=\frac{1}{k_BT}.$$

For a grand-canonical reservoir,

$$\rho_B = \frac{ e^{-\beta(H_B-\mu N_B)} }{ Z_B }.$$

The grand-canonical stationary construction requires $[H_B,N_B]=0$ and finite normalization for the model considered; an infinite reservoir is understood through an appropriate equilibrium limit. See Grand-Canonical Ensemble.

These states are idealizations. Real reservoirs are finite and can heat, cool, saturate, deplete, or develop correlations with the system. The reservoir approximation is controlled when those changes are negligible on the timescale and accuracy of the model.

Noise Source

A noise source is an environment described mainly by its fluctuations. The model may be quantum or classical.

After subtracting mean forces, quantum noise is encoded in ordered correlation functions such as

$$C_{\alpha\beta}(t) = \operatorname{Tr}_B \left[ B_\alpha(t)B_\beta(0)\rho_B \right].$$

For a classical stationary noise process $\xi(t)$, it is encoded in

$$C_{\xi\xi}(t) = \mathbb E[\xi(t)\xi(0)].$$

The spectra obtained by Fourier transforming these correlations determine dephasing rates, transition rates, line broadening, and filter-function estimates. The quantum ordering matters: a quantum bath’s ability to absorb energy need not equal its ability to supply energy. That asymmetry is central to Detailed Balance and the Fluctuation–Dissipation Relation.

Record

An environment can also act as a record. In a decoherence or measurement model, different system alternatives correlate with distinguishable environmental states:

$$\lvert a\rangle\lvert E_0\rangle \longmapsto \lvert a\rangle\lvert E_a\rangle.$$

If the states $\lvert E_a\rangle$ are distinguishable, the environment contains which-alternative information. If many fragments independently contain the same information, those records can become publicly accessible.

This role is different from a thermal reservoir. A photon field may act as a heat bath for an atom, a dephasing environment for a spatial superposition, and a record carrier for a macroscopic object’s position. Which role matters depends on the question being asked.

For the record side, see Environment-Induced Decoherence and Quantum Darwinism Preview.

Comparing the roles

Term	Role	Assumption to state
Environment	Degrees of freedom excluded from the system	Which degrees of freedom are ignored or observed
Bath	Environment represented by a reference state and correlations	Stationarity and negligible change of relevant bath correlations, when used
Reservoir	Source or sink for energy, particles, or another quantity	Which intensive parameters remain approximately fixed
Noise source	Fluctuating force or control parameter	Quantum ordering or classical statistics; spectral convention
Record	Degrees of freedom carrying system information	Distinguishability of the conditional states and accessible readout

The same photon field can play several of these roles. These terms do not by themselves establish equilibrium, short memory, or reliable measurement.

Stationarity and Correlation Time

For a bath state stationary under $H_B$, the correlation function depends on a time difference:

$$C_{\alpha\beta}(\tau) = \operatorname{Tr}_B \left[ B_\alpha(\tau)B_\beta(0)\rho_B \right].$$

The bath correlation time $\tau_B$ is the timescale over which $C_{\alpha\beta}(\tau)$ decays or becomes negligible for the calculation. Markov approximations are usually justified only when $\tau_B$ is short compared with the relevant system timescale $\tau_S$:

$$\tau_B\ll\tau_S.$$

This is a physical condition, not a definition. A bath can be large and stationary but still non-Markovian if its correlations decay slowly, if its spectrum is sharply structured, or if system information returns after being stored in the environment.

See Markov Approximation for the short-memory step and Non-Markovian Dynamics for diagnostics when the environment returns information.

Backaction on the Bath

Bath and reservoir approximations often rely on weak backaction. A heat reservoir remains at approximately the same $T$ after absorbing a small amount of energy. A particle reservoir remains at approximately the same $\mu$ after exchanging a few particles. After an atom emits a photon into a broadband outgoing field, the global field state is not still vacuum. The reservoir approximation instead says that correlations of the field arriving back at the atom remain essentially those of the reference vacuum over the times considered.

This fails when:

• the reservoir is finite;
• the system deposits comparable energy or particles;
• the coupling is strong enough to dress the bath significantly;
• a narrow mode accumulates excitation or repeatedly returns it;
• repeated measurements condition the bath state;
• the bath starts correlated with the system.

When backaction is not negligible, the reservoir should be enlarged into the system, modeled explicitly, or treated with a non-Markovian method. Initial Correlations and Reaction-Coordinate Mapping are common tools for that boundary adjustment.

A stationary bath can retain memory

For one thermal oscillator, take $H_B=\hbar\omega_b b^\dagger b$ and the Hermitian energy-valued bath operator $B=g(b+b^\dagger)$ with real $g$. Its centered correlation is

$$C_{BB}(t)=g^2[(\bar n+1)e^{-i\omega_b t}+\bar n e^{i\omega_b t}], \qquad \bar n=\frac1{e^{\beta\hbar\omega_b}-1}.$$

The state is stationary, but the correlation does not decay. A single undamped mode can exchange excitation coherently with a qubit and cannot be assumed to provide a memoryless reservoir.

Using the explicit convention $S_{BB}(\omega)=\int_{-\infty}^\infty dt\,e^{i\omega t}C_{BB}(t)$ gives peaks of weights $2\pi g^2(\bar n+1)$ at $+\omega_b$ and $2\pi g^2\bar n$ at $-\omega_b$. Their ratio is $e^{-\beta\hbar\omega_b}$. The positive-frequency peak describes the bath’s capacity to absorb system energy; vacuum has no negative-frequency peak for this operator.

For a thermal bath and a Hermitian $B$, the same Fourier convention gives

$$S_{BB}(-\omega)=e^{-\beta\hbar\omega}S_{BB}(\omega),\qquad\omega>0.$$

In a broadband weak-coupling setting, a qubit resonant where $\bar n=1/3$ has excitation-to-decay rate ratio $1/4$, giving stationary excited population $1/5$ when these are its only transitions. A real classical stationary noise process has a symmetric autocorrelation spectrum and gives equal noise-induced up and down rates in this model; additional dissipation is needed to obtain the low-temperature asymmetry. See Detailed Balance and Noise Spectra.

Structured and Engineered Reservoirs

Not all useful reservoirs are featureless continua. A structured reservoir may have a band edge, cavity resonance, finite delay line, pseudomode, phonon peak, or spectral gap. These features can produce memory, nonexponential decay, bound states, or enhanced transition rates.

An engineered reservoir is designed rather than avoided. Examples include:

• sideband cooling reservoirs;
• squeezed electromagnetic reservoirs;
• lossy cavities used for state preparation;
• dissipative stabilization of entangled states;
• reset channels in quantum processors.

Reservoir engineering uses dissipation as a resource. It still requires the same checks: what is the reference state, what correlations matter, how strong is the backaction, and what reduced description is justified?

Examples

Electromagnetic vacuum

For an atom, the vacuum field is a reservoir for photons. It can absorb energy through spontaneous emission even at zero temperature. In a weak-coupling broadband regime, it often supports a Markovian amplitude-damping model.

Phonon bath

Phonons in a solid can act as a thermal bath for spins, excitons, defects, or vibrational modes. Depending on temperature, dimensionality, and spectral structure, the bath can produce relaxation, dephasing, or non-Markovian memory.

Oscillator bath

The Caldeira–Leggett model represents an environment by many harmonic oscillators. It is a useful idealization for damping, Brownian motion, and force noise, but its predictions depend strongly on the spectral density and counterterms.

Spin bath

A finite collection of nuclear spins or paramagnetic impurities can be an environment without being a thermal oscillator bath. Spin baths often have long memory, non-Gaussian statistics, and strong finite-size effects.

Classical noise environment

Slow fluctuations in a control field may be modeled as a classical random process. This can be appropriate when the noise source is effectively macroscopic and backaction from the system is negligible. It cannot capture quantum absorption and emission asymmetry by itself.

Measurement record

Photons scattered from a pointer, current in an amplifier, or states of ancillas in a measurement circuit can be environmental records. If ignored, they produce decoherence; if observed, they support conditional trajectories or measurement outcomes.

Exercises

1. Classify the outside degrees of freedom. For each case, say whether the outside degrees of freedom are best described as an environment, bath, reservoir, noise source, record, or more than one of these:

   • a single ancilla qubit that interacts once and is later measured;
   • a broadband electromagnetic vacuum coupled to an excited atom;
   • low-frequency flux noise in a superconducting qubit;
   • photons scattered from a macroscopic pointer;
   • a finite cavity mode strongly coupled to a qubit.

Solution

The ancilla is an environment and, if its conditional states are distinguishable, already a record before it is read; it is not usually a bath because it is a single controlled degree of freedom.

The electromagnetic vacuum is an environment and often a reservoir for photons and energy. In broadband weak-coupling regimes it is also treated as a bath.

Low-frequency flux noise is commonly modeled as a noise source. If a microscopic model is specified, it may be a bath; if it is treated as a fitted classical stochastic process, the noise-source description is the safer one.

Scattered photons are environment degrees of freedom and records of pointer position. They can also contribute to decoherence.

A strongly coupled finite cavity mode is an environment if excluded from the system, but it may be better included in an enlarged system. It is not a featureless bath unless further damping modes are added.

2. Stationary correlation. Let $\rho_B$ commute with $H_B$, and let

$$B(t) = e^{iH_Bt/\hbar}B e^{-iH_Bt/\hbar}.$$

Show that

$$\operatorname{Tr}_B[B(t)B(s)\rho_B]$$

depends only on $t-s$.

Solution

Use

$$B(t) = e^{iH_Bs/\hbar} B(t-s) e^{-iH_Bs/\hbar},$$

and

$$B(s) = e^{iH_Bs/\hbar} B e^{-iH_Bs/\hbar}.$$

Then

$$B(t)B(s) = e^{iH_Bs/\hbar} B(t-s)B e^{-iH_Bs/\hbar}.$$

Therefore

$$\operatorname{Tr}_B[B(t)B(s)\rho_B] = \operatorname{Tr}_B \left[ B(t-s)B e^{-iH_Bs/\hbar}\rho_B e^{iH_Bs/\hbar} \right].$$

Since $[H_B,\rho_B]=0$, the conjugated $\rho_B$ is just $\rho_B$. Thus the correlation is

$$\operatorname{Tr}_B[B(t-s)B\rho_B],$$

which depends only on $t-s$.

3. Markov diagnostic. A bath correlation decays on $\tau_B=0.2\,\mathrm{ps}$, while the system population changes on $\tau_S=5\,\mathrm{ns}$. Is the short-memory part of a Markov approximation plausible? What else must still be checked?

Solution

The ratio is

$$\frac{\tau_B}{\tau_S} = \frac{0.2\,\mathrm{ps}}{5\,\mathrm{ns}} = 4\times10^{-5},$$

so the separation of correlation and system timescales is strong. That supports the short-memory part of a Markov approximation.

One must still check weak coupling, bath stationarity, absence of important initial correlations, whether the relevant system frequencies sample smooth bath spectra, and whether a secular or coarse-graining step is needed for complete positivity.

4. Reservoir backaction. A qubit repeatedly dumps energy $\hbar\omega_0$ into a small resonator containing only a few photons. Why is it risky to call the resonator a fixed-temperature reservoir?

Solution

A fixed-temperature reservoir should absorb energy without appreciably changing its state. A small resonator has few degrees of freedom, so added energy changes its occupation and later dynamics. It can feed energy back to the qubit or produce revivals. An ideal harmonic resonator has no finite occupation saturation, although its occupation and transition rates can change substantially. The resonator should likely be modeled explicitly as part of an enlarged system, with any remaining damping modes treated as the reservoir.

5. Classical versus quantum spectra. Why can a real classical noise process describe dephasing well but fail to model low-temperature relaxation correctly?

Solution

A real classical stationary noise process has a symmetric spectrum: positive and negative frequencies have equal weight. A quantum thermal bath has ordered spectra whose positive and negative frequency components differ by detailed balance. Low-temperature relaxation depends on the bath absorbing energy more readily than supplying it. Classical symmetric noise cannot encode that absorption-emission asymmetry without additional modeling.

References

• H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford University Press (2002). Publisher.
• C. W. Gardiner and P. Zoller, Quantum Noise, 3rd ed., Springer (2004). Publisher.
• Á. Rivas and S. F. Huelga, Open Quantum Systems: An Introduction, Springer (2012). Author manuscript.
• G. Schaller, Open Quantum Systems Far from Equilibrium, Springer (2014).
• U. Weiss, Quantum Dissipative Systems, 4th ed., World Scientific (2012).