Quantum Noise

Quantum noise is the effect of uncontrolled degrees of freedom on the system being modeled. It includes ordinary classical randomness, but it is broader: a quantum environment has operators, noncommuting fluctuations, vacuum fluctuations, asymmetric spectra, and backaction. Even a zero-temperature electromagnetic field is not noiseless for a system that can emit into it.

The central distinction is:

classical noise is a random number or function;
quantum noise is usually an operator of unobserved degrees of freedom.

This distinction matters for relaxation rates, detailed balance, measurement limits, amplifier noise, spontaneous emission, dephasing, and thermodynamic consistency.

For the terminology separating environments, baths, reservoirs, noise sources, and records, see Baths, Reservoirs, and Environments. For frequency-domain correlations, see Noise Spectra.

Noise as a Modeling Boundary

Noise is not an intrinsic label attached to a physical object. It is a role assigned after choosing a system boundary. A microwave transmission line can be:

• part of the environment when its outgoing field is ignored;
• a bath when it remains close to a stationary reference state;
• a reservoir when it supplies or absorbs photons;
• a measurement record when its output is monitored;
• a control resource when its field is shaped deliberately.

The same physical modes can therefore appear as noise, signal, reservoir, record, or controller in different models. A good open-system description states which degrees of freedom are retained, which are traced out, and which are measured.

Classical Stochastic Noise

Classical stochastic noise is represented by a random c-number process, such as $\xi(t)$. A simple Hamiltonian model is

$$H(t) = H_0 + A\,\xi(t),$$

where $A$ is a system operator and $\xi(t)$ is a prescribed random process. The system state for one realization evolves unitarily. The noise-averaged state is obtained by averaging over realizations:

$$\rho(t) = \mathbb E_{\xi} \!\left[ U_\xi(t)\rho(0)U_\xi^\dagger(t) \right].$$

This model is useful for technical fluctuations such as magnetic-field drift, laser phase noise, voltage noise, oscillator phase noise, and slow disorder. It is also often used as an effective description of a high-temperature or strongly measured environment.

Classical noise is not always harmless. It can be non-Gaussian, nonstationary, heavy-tailed, correlated with controls, or dependent on the system through feedback. The phrase “add classical noise” does not by itself define a valid model.

Quantum Bath Noise

In a microscopic open-system model, the fluctuating quantity is usually a bath operator. A standard coupling is

$$H_{\mathrm{int}} = A\otimes B,$$

where $A$ acts on the system and $B$ acts on the bath. In the interaction picture, the bath correlation

$$C_{BB}(t) = \langle B(t)B(0)\rangle$$

controls weak-coupling transition rates and memory effects.

The operator order matters. In general,

$$\langle B(t)B(0)\rangle \ne \langle B(0)B(t)\rangle.$$

That asymmetry has physical content. A quantum bath may absorb energy from the system differently than it supplies energy to the system. This is why the unsymmetrized spectrum, not only the symmetrized noise power, appears in many master-equation rates.

Vacuum Noise

Vacuum is not the absence of quantum noise. A field mode in its ground state has no real photons on average, but it still has nonzero fluctuations and commutation relations.

For a broadband bosonic input field in vacuum, a common Markovian convention is

$$\langle b_{\mathrm{in}}(t)b_{\mathrm{in}}^\dagger(t')\rangle = \delta(t-t'), \qquad \langle b_{\mathrm{in}}^\dagger(t)b_{\mathrm{in}}(t')\rangle = 0.$$

The first correlation allows an excited system to emit into the vacuum. The second says the vacuum does not thermally excite the system. This asymmetry is the noise-language version of spontaneous emission without thermal absorption.

Vacuum noise is central in quantum optics, circuit QED, quantum-limited amplification, radiation pressure noise, and homodyne shot noise. For the traveling-field normalization, see Input–Output Theory.

Thermal Noise

Thermal quantum noise comes from a bath in a thermal state

$$\rho_B = \frac{e^{-\beta H_B}}{Z}.$$

Thermal noise contains both quantum zero-point fluctuations and occupation-dependent fluctuations. For a bosonic mode at frequency $\omega$,

$$\bar n(\omega) = \frac{1}{e^{\beta\hbar\omega}-1}.$$

In a broadband input model,

$$\langle b_{\mathrm{in}}^\dagger(t)b_{\mathrm{in}}(t')\rangle = \bar n\,\delta(t-t'),$$

and

$$\langle b_{\mathrm{in}}(t)b_{\mathrm{in}}^\dagger(t')\rangle = (\bar n+1)\delta(t-t').$$

The $+1$ term is the vacuum contribution. At high temperature, $\bar n$ is large and the noise often looks approximately classical. At low temperature, the asymmetry between emission and absorption is essential.

For equilibrium constraints on thermal spectra, see Fluctuation–Dissipation Relation and Detailed Balance.

Technical Noise

Technical noise is noise from imperfect apparatus rather than from an ideal equilibrium bath. Examples include:

• laser phase and intensity noise;
• microwave source noise;
• amplifier gain fluctuations;
• flux, charge, or gate-voltage drift;
• acoustic and vibrational noise;
• thermal gradients and fluctuating backgrounds;
• digitizer noise and calibration drift;
• slow $1/f$ noise in solid-state devices.

Technical noise is often effectively classical over the bandwidth of interest, but that is an approximation. A noisy amplifier or transmission line may require quantum input–output modeling near the quantum limit, while slow drift in a control parameter may be better treated as a classical stochastic process.

The practical question is not whether the noise is “really classical” in some absolute sense. The practical question is which model predicts the observed correlations, backaction, and response within the required accuracy.

Noise Versus Measurement Record

Noise and record are complementary roles. If environmental degrees of freedom carry away information and are ignored, they appear as noise or decoherence. If those degrees of freedom are measured and retained, they become a record and can condition the state.

For example, an emitted optical field can be:

• ignored, producing an unconditional decay channel;
• photon counted, producing a jump record;
• homodyne measured, producing a diffusive quadrature record;
• fed into a controller, producing measurement-based feedback.

The underlying coupling may be the same. What changes is which information is kept. For continuous records, see Measurement Records and Quantum Filtering.

Correlations and Spectra

Most open-system effects depend not on the instantaneous noise value but on correlations. For a stationary bath operator $B(t)$,

$$C_{BB}(t) = \langle B(t)B(0)\rangle, \qquad S_{BB}(\omega) = \int_{-\infty}^{\infty} dt\,e^{i\omega t}C_{BB}(t).$$

The correlation time tells how quickly the bath forgets:

$$\tau_B \sim \text{width of }C_{BB}(t).$$

The Markov approximation is plausible only when the bath correlation time is short compared with the system evolution being modeled. Long-tailed correlations, narrow spectral features, strong coupling, or delayed feedback require more careful non-Markovian treatment.

For the frequency-domain conventions and physical interpretation of positive and negative frequencies, see Noise Spectra.

White Noise and Markov Noise

White noise is an idealization in which the correlation time is taken to zero while the integrated noise strength remains finite. In time domain, a classical white-noise model has

$$\mathbb E[\xi(t)\xi(t')] = D\,\delta(t-t').$$

Quantum Markov input fields have analogous delta correlations, but also nontrivial commutators. For a bosonic input,

$$[b_{\mathrm{in}}(t),b_{\mathrm{in}}^\dagger(t')] = \delta(t-t').$$

This commutator is not optional bookkeeping. It is what preserves system commutation relations in quantum Langevin equations and enforces quantum noise limits in amplifiers.

White noise is never literally valid at all frequencies. It is an effective model over the bandwidth relevant to the system and detector. When a high-frequency cutoff, finite detector bandwidth, or structured density of states matters, the white-noise approximation should be replaced by a finite-bandwidth model.

What Noise Does to a System

Different parts of the noise spectrum affect different physical processes.

Low-frequency longitudinal fluctuations produce dephasing. For a qubit with Hamiltonian perturbation

$$H_{\xi}(t) = \frac{\hbar}{2}\xi(t)\sigma_z,$$

slow noise changes the relative phase between $\sigma_z$ eigenstates. Echo and dynamical-decoupling sequences suppress part of this noise by changing the filter function.

Noise near a transition frequency produces relaxation or excitation. If a system operator $A$ couples two energy eigenstates separated by $\hbar\omega_0$, then bath spectral weight near $\pm\omega_0$ controls upward and downward transition rates.

Noise coupled to position or momentum can produce diffusion, damping, and heating. The Caldeira–Leggett Model is the standard oscillator-bath example.

Measurement noise appears as imprecision and backaction. A detector record contains useful signal plus fluctuations, while the measurement interaction also disturbs the system. Quantum-limited measurement requires accounting for both.

Classical Approximation to Quantum Noise

A quantum bath can sometimes be replaced by classical stochastic noise, but this is a controlled approximation, not an identity. It is safest when:

• the relevant bath observables commute approximately over the timescales of interest;
• the bath temperature is high compared with relevant transition energies;
• only symmetrized fluctuations matter;
• backaction of the system on the bath response is negligible;
• detailed balance and absorption-emission asymmetry are unimportant.

The approximation is unsafe when spontaneous emission, low-temperature absorption, quantum-limited amplification, noncommuting quadratures, or measurement backaction are central.

The diagnostic question is:

$$\text{Does the model need ordered operator correlations?}$$

If yes, a purely classical noise model is probably missing essential physics.

Common Mistakes

• Treating vacuum as no noise.
• Using a symmetrized spectrum when a transition rate needs an ordered spectrum.
• Assuming all noise spectra are even functions of frequency.
• Calling a bath “thermal” without checking detailed balance.
• Applying white-noise formulas outside their bandwidth of validity.
• Replacing a quantum bath by classical noise at low temperature without checking emission and absorption asymmetry.
• Confusing an unobserved noise channel with a measurement record.
• Ignoring technical noise because a microscopic bath model looks clean.
• Treating $1/f$ noise as stationary over arbitrarily long times without specifying cutoffs.

Cross-Links

• Baths, Reservoirs, and Environments for terminology and modeling boundaries.
• System–Bath Hamiltonians for microscopic couplings.
• Correlation Functions for time-domain bath correlations and correlation times.
• Noise Spectra for frequency-domain correlations.
• Fluctuation–Dissipation Relation for equilibrium noise-response constraints.
• Thermal and Vacuum Noise for occupation factors, spontaneous emission, and thermal excitation.
• Quantum Langevin Equations for Heisenberg-picture damping with input noise operators.
• Input–Output Theory for Markovian traveling fields.
• Pure Dephasing Model and Pure Dephasing Master Equation for phase-noise consequences.
• One-Over-F Noise for cutoff-dependent low-frequency noise and solid-state dephasing.
• Amplitude Damping Master Equation for relaxation consequences.
• Josephson Effect for the junction phase, RCSJ environment, stochastic switching, and the boundary between a white-noise source and frequency-dependent circuit impedance.
• Measurement Records for the distinction between unobserved noise and retained records.
• Radiation Pressure for photon-recoil force noise and momentum diffusion.
• Approximation Checklist for Markov, thermal, secular, and positivity checks.

References

• C. W. Gardiner and P. Zoller, Quantum Noise, 3rd ed., Springer, 2004.
• D. F. Walls and G. J. Milburn, Quantum Optics, 2nd ed., Springer, 2008.
• A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Marquardt, and R. J. Schoelkopf, “Introduction to quantum noise, measurement, and amplification,” Reviews of Modern Physics 82, 1155–1208, 2010.
• H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford University Press, 2002.
• U. Weiss, Quantum Dissipative Systems, 4th ed., World Scientific, 2012.
• H. M. Wiseman and G. J. Milburn, Quantum Measurement and Control, Cambridge University Press, 2010.

Exercises

1. Explain why vacuum can cause spontaneous emission but not thermal excitation of a two-level system.

Solution

For a vacuum input field,

$$\langle b_{\mathrm{in}}(t)b_{\mathrm{in}}^\dagger(t')\rangle = \delta(t-t'), \qquad \langle b_{\mathrm{in}}^\dagger(t)b_{\mathrm{in}}(t')\rangle = 0.$$

The first correlation corresponds to the field accepting an emitted quantum from the system. The second would correspond to photons already present in the field exciting the system, and it vanishes in vacuum. Thus vacuum supports emission but not thermal absorption.

2. A real classical stationary noise process has an even two-sided spectrum. Why is this not generally true for a quantum ordered spectrum?

Solution

For a classical process, ordinary multiplication commutes, and the correlation function for a real stationary process satisfies $C(t)=C(-t)$ under the usual assumptions. The spectrum is therefore even. For a quantum bath, the ordered correlation $\langle B(t)B(0)\rangle$ is not generally equal to $\langle B(0)B(t)\rangle$ because the operators need not commute. The positive- and negative-frequency parts can therefore differ, encoding absorption-emission asymmetry.

3. Give one reason why slow technical noise is often modeled classically and one reason why this can fail.

Solution

Slow technical noise is often modeled classically because it may come from macroscopic apparatus drift, such as voltage, magnetic-field, or laser-frequency fluctuations, whose measured values can be treated as ordinary random variables over the system bandwidth. This can fail if the noise source is near the quantum limit, has non-negligible backaction, is correlated with the system through feedback, or affects transitions where quantum absorption-emission asymmetry matters.

4. In the coupling $H_{\mathrm{int}}=A\otimes B$, why do transition rates depend on bath correlations rather than on an instantaneous value of $B$?

Solution

Transitions are produced by bath fluctuations at frequencies matching system energy gaps. Those frequency components are determined by two-time correlations such as $\langle B(t)B(0)\rangle$ and their Fourier transforms. An instantaneous expectation value of $B$ can often be absorbed into a mean field or Hamiltonian shift; the fluctuating correlations determine irreversible rates and noise-induced dynamics.