The Kubo Formula

How fast does charge move through a metal, heat through an insulator, momentum through a fluid? These transport coefficients — electrical and thermal conductivity, viscosity, diffusion — quantify how a system responds to a weak drive, and they seem to demand that you actually drive the system out of equilibrium and watch it flow. The Kubo formula, written down by Ryogo Kubo in 1957, says otherwise. Each transport coefficient is exactly a time integral of an equilibrium correlation function of the relevant flux. No field appears in the final expression: the system's own spontaneous thermal jiggling — the way its currents fluctuate and decay in equilibrium — already contains everything about how it will dissipate energy when perturbed. Watch the fluctuations, integrate, and you have the conductivity. The derivation is a targeted application of linear response theory. Add a weak time-dependent coupling $H'(t)=-\sum_j A_j F_j(t)$ between observables and external fields, solve the Liouville–von Neumann equation for the density matrix to first order, and the induced change of any observable $B$ becomes a causal convolution of the field with a retarded kernel built from equilibrium commutators (quantum) or Poisson brackets (classical). That kernel is the Kubo response function $\chi_{BA}(t)=\tfrac{i}{\hbar}\Theta(t)\langle[B(t),A(0)]\rangle_0$ . Taking $B$ and $A$ to be currents and passing to the DC limit yields the Green–Kubo relation, the exact microscopic expression for a transport coefficient as $\int_0^\infty\langle J(0)J(t)\rangle\,\mathrm{d}t$ . Two structural features give the result its authority. Causality forces $\chi(t)$ to vanish for $t<0$ , tying its real and imaginary parts through the Kramers–Kronig relations. And the fluctuation-dissipation theorem relates the dissipative $\chi''(\omega)$ to the equilibrium fluctuation spectrum, so absorption at frequency $\omega$ is governed by spontaneous fluctuations at that frequency. In the classical, low-frequency limit the quantum commutators become current–current time correlations — the Green–Kubo formulas now used routinely in molecular dynamics to extract diffusion constants, viscosities, and conductivities from equilibrium simulations. Onsager's reciprocal relations fall out as a consequence of the microscopic time-reversal symmetry of the same correlators.

Key takeaways

The Kubo formula expresses linear-response transport coefficients and susceptibilities exactly as time integrals of equilibrium correlation functions of the conjugate fluxes — no driving appears in the final expression.
It follows from first-order time-dependent perturbation of the density matrix for a Hamiltonian perturbed by $-\sum_j A_j F_j(t)$ .
The central object is the retarded response function $\chi_{BA}(t)=\tfrac{i}{\hbar}\Theta(t)\langle[B(t),A(0)]\rangle_0$ , whose Fourier transform is the dynamic susceptibility $\chi(\omega)$ .
For DC conductivity it gives the Green–Kubo relation $\sigma=\dfrac{1}{Vk_BT}\int_0^\infty\langle J(t)J(0)\rangle\,\mathrm{d}t$ , and for diffusion $D=\int_0^\infty\langle v(t)v(0)\rangle\,\mathrm{d}t$ , linking transport to flux fluctuations.
Its classical limit recovers all the Green–Kubo formulas (diffusion, viscosity, thermal conductivity), and its structure delivers the Einstein relation, Nyquist noise, and Onsager reciprocity.
Key idea: couple the system to a weak field via $H'=-A\,F(t)$ , expand the density matrix to first order, and the induced response of any $B$ becomes an equilibrium time-correlation function — transport equals integrated fluctuations.

The core insight

Couple the system to a weak field through $H'=-A\,F(t)$ , expand the density matrix to first order, and the induced response of any $B$ becomes an equilibrium time-correlation function — transport equals integrated fluctuations.

Where this lives today

The Green–Kubo route is the default way to compute transport coefficients from atomistic simulation. Record the equilibrium autocorrelation of the appropriate flux — charge current for ionic and electronic conductivity, stress tensor for shear viscosity, heat current for thermal conductivity, tagged-particle velocity for self-diffusion — and integrate. Because no external field is applied, the method avoids the finite-size and nonlinearity artifacts that plague explicitly driven (nonequilibrium) molecular dynamics, and it is the workhorse for electrolytes, molten salts, and battery materials where transport is the property of interest. In condensed-matter theory the Kubo formula is the starting point for the entire linear-response program: optical conductivities and their f-sum rules, the anomalous and spin Hall conductivities (where the reactive part carries a Berry-curvature contribution), and the Kubo–Greenwood formula used in first-principles electronic-structure codes to compute conductivities and thermopowers from Kohn–Sham states. Wherever a transport coefficient is written as an integral over an equilibrium correlation function — from lattice thermal conductivity to the shear viscosity of the quark–gluon plasma — the object being invoked is the Kubo formula.

The physical picture

Gently shake a system with a weak external field and the Kubo formula says it "already knows" how it will react: its equilibrium thermal jiggling — the way spontaneous currents fluctuate and decay — encodes the entire response. You never need to apply a strong field and measure the steady flow; you just record the fluctuations at equilibrium and integrate them. A metal's DC resistance is, quite literally, set by how quickly spontaneous current fluctuations relax back to zero: faster decay (more scattering) means a smaller integrated correlation, hence lower conductivity. Causality is again the organizing principle. Because a response cannot precede its cause, $\chi(t)=0$ for $t<0$ ; applied to the analytic structure of the Fourier transform this links the in-phase reactive response ( $\mathrm{Re}\,\chi$ ) and the out-of-phase dissipative response ( $\mathrm{Im}\,\chi$ ) through the Kramers–Kronig relations. The commutator $[B(t),A(0)]$ is the quantum heart of the formula; classically it becomes a Poisson bracket, and the $\hbar\to0$ limit $\tfrac{i}{\hbar}[\,,\,]\to\{\,,\,\}$ delivers the classical Green–Kubo relations from the same master expression.

Figure

The velocity autocorrelation and the diffusion constant: D is the area under the curve — a dense-liquid curve with a negative back-scattering dip integrates to a SMALLER D, the outlier where correlations suppress transport

The DC conductivity is a long-time integral of the current–current correlation. That integral is delicate: a genuinely conducting system has a correlation that decays to zero, so the integral converges; a superconductor or a ballistic system carries a non-decaying "Drude weight" (a $\delta$ -function at $\omega=0$ ) that is not ordinary dissipative transport. The order of limits — thermodynamic limit $V\to\infty$ before $\omega\to0$ — is what separates real dissipation from these singular pieces.

Figure

The Kubo pipeline: a weak field perturbs the density matrix to first order, the trace becomes an equilibrium commutator, and the DC limit collapses it to a Green-Kubo current integral

The core equations

The perturbation couples observables to weak generalized forces. Everything downstream is a first-order functional of this coupling and the equilibrium ensemble.

H(t)=H_0-\sum_j A_j\,F_j(t) \qquad (1)

where: $H_0$ — unperturbed (equilibrium) Hamiltonian; $A_j$ — observable coupling to the external field; $F_j(t)$ — weak time-dependent generalized force. Physically, $A_j$ is the "handle" (polarization for an electric field, position for a mechanical force) and the observed $B$ is the induced flux. The induced change of any observable is a causal convolution with a retarded response function — the general shape of every linear-response statement.

\delta\langle B(t)\rangle=\sum_j\int_{-\infty}^{t}\chi_{BA_j}(t-t')\,F_j(t')\,\mathrm{d}t' \qquad (2)

where: $\delta\langle B\rangle$ — field-induced change in $B$ ; $\chi_{BA}$ — retarded (after-effect) response function. The past-only integration limit is causality; the response follows the force, never precedes it. The Kubo response function is the equilibrium average of a commutator of Heisenberg operators. This is the microscopic expression that names the whole framework.

\chi_{BA}(t)=\frac{i}{\hbar}\,\Theta(t)\,\big\langle[B(t),A(0)]\big\rangle_0 \qquad (3)

where: $\Theta(t)$ — Heaviside step (causality); $B(t)=e^{iH_0t/\hbar}B\,e^{-iH_0t/\hbar}$ — Heisenberg operator under $H_0$ ; $\langle\cdots\rangle_0$ — average in the unperturbed equilibrium ensemble. The commutator measures how $B$ at time $t$ fails to commute with $A$ at time $0$ — the propagated, time-resolved response. The DC conductivity is the celebrated Green–Kubo relation: a transport coefficient equal to the time-integral of an equilibrium current–current correlation, with $1/k_BT$ the only prefactor.

\sigma_{\mu\nu}^{\,\mathrm{DC}}=\frac{1}{V k_BT}\int_{0}^{\infty}\big\langle J_\mu(t)\,J_\nu(0)\big\rangle_0\,\mathrm{d}t \qquad (4)

where: $J_\mu$ — $\mu$ -component of the total current; $V$ — system volume; $k_BT$ — thermal energy. Faster-decaying current correlations (more scattering) give a smaller integral and lower conductivity — resistance is the rate at which currents forget themselves. The single-particle instance is the diffusion constant as the integral of the velocity autocorrelation, which together with the mobility gives the Einstein relation.

D=\int_0^\infty\big\langle v(t)\,v(0)\big\rangle_0\,\mathrm{d}t,\qquad \mu=\frac{D}{k_BT} \qquad (5)

where: $D$ — self-diffusion constant; $\mu$ — mobility (drift velocity per unit force). The equality $D=\mu k_BT$ is the fluctuation ( $D$ ) and dissipation ( $\mu^{-1}$ ) sides of one theorem — the same content as the Einstein relation, here delivered by the Kubo integral. The dissipative part of the response is tied to the fluctuation spectrum by the fluctuation-dissipation theorem, the bridge from Kubo response to equilibrium noise.

\mathrm{Im}\,\chi_{AA}(\omega)=\frac{1}{2\hbar}\big(1-e^{-\beta\hbar\omega}\big)\,S_{AA}(\omega) \qquad (6)

where: $S_{AA}(\omega)=\int e^{i\omega t}\langle A(t)A(0)\rangle\,\mathrm{d}t$ — fluctuation power spectrum; the factor $(1-e^{-\beta\hbar\omega})$ is the KMS detailed-balance weight. Classically it reduces to $\mathrm{Im}\,\chi=\tfrac{\beta\omega}{2}S$ , i.e. $S=(2k_BT/\omega)\mathrm{Im}\,\chi$ . Causality forces the real (reactive) part to be a principal-value transform of the imaginary (dissipative) part — the Kramers–Kronig relation, used to enforce sum rules on measured spectra.

\mathrm{Re}\,\chi(\omega)=\frac{1}{\pi}\,\mathcal{P}\!\!\int_{-\infty}^{\infty}\frac{\mathrm{Im}\,\chi(\omega')}{\omega'-\omega}\,\mathrm{d}\omega' \qquad (7)

where: $\mathcal{P}$ — Cauchy principal value. It is the analytic fingerprint of $\chi(t)=0$ for $t<0$ .

The core equations continued — Onsager reciprocity

Microscopic time-reversal symmetry of the equilibrium correlators makes the matrix of transport coefficients symmetric (in zero field), a result Onsager derived and Kubo's framework makes transparent.

L_{ij}(B)=L_{ji}(-B) \qquad (8)

where: $L_{ij}$ — matrix of kinetic (transport) coefficients coupling fluxes and forces; $B$ — magnetic field, which flips sign under time reversal. In zero field $L_{ij}=L_{ji}$ : for instance the Peltier coefficient equals $T$ times the Seebeck coefficient (the Kelvin relation $\Pi=TS$ ).

Derivation AThe Kubo formula for a transport coefficient from the current-current correlation

The goal is to turn the retarded Kubo kernel — whose density-matrix derivation is OWNED by linear response theory (recalled by name, not re-derived here) — into a transport coefficient, showing that a dissipative response measured at low frequency equals a DC integral of an equilibrium current-current correlation. We take the equilibrium kernel $\chi_{BA}(t)=\tfrac{i}{\hbar}\Theta(t)\langle[B(t),A(0)]\rangle_0$ as established, specialize the observable to a current, pass to the frequency domain, and extract the $\omega\to0$ dissipative part. The payoff is the Green–Kubo formula: a transport coefficient is the time-integral of an equilibrium flux autocorrelation, with no external field in the final expression.

Step 1. Take the classical Kubo kernel from linear response theory: for a perturbation $H'=-A\,F(t)$ , the response of an observable $B$ is $\chi_{BA}(t)=-\beta\,\Theta(t)\,\langle\dot A(0)\,B(t)\rangle_0$ , an equilibrium average. Why: linear response theory owns this kernel; the Kubo formula only specializes it, taking a current as the observable.

Step 2. Fix the flux geometry: let the force couple to a coordinate whose velocity is the current $J$ , so $B=J$ and $\dot A=J$ . The kernel collapses to $\chi_{JJ}(t)=\beta\,\Theta(t)\,\langle J(0)\,J(t)\rangle_0$ . Why: transport is the response of a current to its conjugate field, so both operator slots carry the same current $J$ — the object that fluctuates in equilibrium.

Step 3. Take the transport (DC) limit: $L=\lim_{\omega\to0}\tfrac{1}{\omega}\,\chi''_{JJ}(\omega)=\beta\int_0^\infty\langle J(0)\,J(t)\rangle_0\,\mathrm{d}t$ . Why: the dissipative part at low frequency is exactly the running integral of the flux autocorrelation — the Green–Kubo identity, turning a driven-transport measurement into an equilibrium fluctuation.

Step 4. Read off the standard instances: diffusion $D=\int_0^\infty\langle v(0)v(t)\rangle\,\mathrm{d}t$ and conductivity $\sigma=\tfrac{1}{Vk_BT}\int_0^\infty\langle J(0)J(t)\rangle\,\mathrm{d}t$ . Why: every transport coefficient is one flux autocorrelation integrated to zero frequency — the entire content of the Kubo formula. *Verified: verify_aq-sm-kubo-formula.py confirms the classical kernel $\chi=-\beta\Theta\,\langle\dot A B\rangle$ (used, not re-derived), the current–current reduction $L=\beta\int_0^\infty\langle JJ\rangle$ , and the static limit $\chi(0)=\beta\,\mathrm{Var}(A)$ .*

Go deeper · Derivation B · 1 minVelocity autocorrelation, Green-Kubo diffusion, and the Einstein relation

Derivation BVelocity autocorrelation, Green-Kubo diffusion, and the Einstein relation

The goal is to derive the Green–Kubo expression $D=\int_0^\infty\langle v(t)v(0)\rangle\,\mathrm{d}t$ for the diffusion constant and recover the Einstein relation $D=\mu k_BT$ , using the exponential velocity autocorrelation of the Langevin/Ornstein–Uhlenbeck model as the concrete check. We couple position to a force, apply the classical Kubo kernel with velocity as the observed flux, integrate to get the static mobility, and identify the integrated correlation as the diffusion constant. The strategy exposes why a transport coefficient really is an integrated equilibrium fluctuation.

Step 1. Couple $A=x$ (position) to a force $F(t)$ and observe $B=v=\dot x$ ; the induced drift is $\delta\langle v(t)\rangle=\int_{-\infty}^t\chi_{vx}(t-t')F(t')\,\mathrm{d}t'$ . Why: mobility is the zero-frequency velocity response to a constant force, so $x$ is the natural conjugate coupling.

Step 2. The classical Kubo kernel is $\chi_{vx}(t)=\beta\Theta(t)\langle\dot v(t)v(0)\rangle_0=-\beta\Theta(t)\tfrac{\mathrm{d}}{\mathrm{d}t}\langle v(t)v(0)\rangle_0$ , using $\dot x=v$ and stationarity $\langle\dot v(t)v(0)\rangle=-\tfrac{\mathrm{d}}{\mathrm{d}t}\langle v(t)v(0)\rangle$ . Why: writing the response as a total time derivative of the correlation makes the DC integral telescope to boundary values.

Step 3. The static mobility is $\mu=\int_0^\infty\chi_{vx}(t)\,\mathrm{d}t=\beta[\langle v(0)^2\rangle-\langle v(0)v(\infty)\rangle]=\beta\langle v^2\rangle$ (the correlation decays to zero); with equipartition $\langle v^2\rangle=k_BT/m$ this is $\mu=1/(m\gamma)$ for the exponential model. Why: integrating the total derivative gives the difference of endpoint correlations, and only the $t=0$ value $\langle v^2\rangle$ survives.

Step 4. Identify $D=\int_0^\infty\langle v(t)v(0)\rangle\,\mathrm{d}t$ and combine: $D=\mu k_BT$ . For the OU/Langevin autocorrelation $\langle v(t)v(0)\rangle=(k_BT/m)e^{-\gamma t/m}$ , the integral gives $D=(k_BT/m)(m/\gamma)=k_BT/(m\gamma)$ and $\mu=1/(m\gamma)$ , the Stokes result. Why: this is the concrete confirmation — the same exponential correlation that defines the <strong>Ornstein-Uhlenbeck process</strong> integrates to the diffusion constant, tying Green–Kubo to Einstein. *Verified: verify_aq-sm-kubo-formula.py integrates the exponential velocity autocorrelation numerically to get $D=k_BT/(m\gamma)$ , confirms $\mu=1/(m\gamma)$ , $D=\mu k_BT$ , and the single-carrier conductivity $\sigma=nq^2D/k_BT=nq^2\mu$ .*

Limiting cases

The Kubo formula specializes to familiar results in the key regimes; each row is a recognizable transport law.

Case	Result	Meaning
Classical limit	$\tfrac{i}{\hbar}\langle[B(t),A]\rangle\xrightarrow{\hbar\to0}\beta\langle\dot B(t)A(0)\rangle$	The commutator becomes a $\beta$ -weighted time-correlation; yields the classical Green–Kubo relations.
DC conductivity	$\sigma_{\mu\nu}=\dfrac{1}{Vk_BT}\int_0^\infty\langle J_\mu(t)J_\nu(0)\rangle\,\mathrm{d}t$	Static ( $\omega\to0$ ) limit of the optical conductivity, valid after $V\to\infty$ .
Diffusion (Einstein)	$D=\dfrac{1}{d}\int_0^\infty\langle\mathbf{v}(t)\cdot\mathbf{v}(0)\rangle\,\mathrm{d}t$	Single-particle velocity autocorrelation; with mobility gives $D=\mu k_BT$ .
Static susceptibility	$\chi_{BA}(0)=\beta[\langle BA\rangle-\langle B\rangle\langle A\rangle]$	Zero-frequency response is an equal-time equilibrium covariance — the classical Kubo identity.

Worked examples

Example 1 — Diffusion constant from the velocity autocorrelation, and the mobility that follows. This turns a decaying correlation into a real transport number and checks the Einstein relation.

Step 1. Take a Brownian particle with the Langevin velocity autocorrelation $\langle v(t)v(0)\rangle=(k_BT/m)e^{-\gamma t/m}$ , with $m=1.0\times10^{-15}\,\mathrm{kg}$ (a micron-scale bead), friction ratio $\gamma/m=1.0\times10^{6}\,\mathrm{s^{-1}}$ , and $T=300\,\mathrm{K}$ .

Step 2. Integrate: $D=\int_0^\infty(k_BT/m)e^{-\gamma t/m}\mathrm{d}t=(k_BT/m)\cdot(m/\gamma)=k_BT/\gamma$ . Note the mass cancels, leaving $D=k_BT/\gamma$ with $\gamma=m(\gamma/m)=(10^{-15})(10^6)=1.0\times10^{-9}\,\mathrm{kg/s}$ .

Step 3. Evaluate: $D=(1.381\times10^{-23})(300)/(1.0\times10^{-9})=4.143\times10^{-21}/10^{-9}=4.143\times10^{-12}\,\mathrm{m^2/s}$ .

Step 4. Get the mobility from Einstein: $\mu=D/k_BT=1/\gamma=1.0\times10^{9}\,\mathrm{s/kg}$ . The bead diffuses at $D\approx4.14\times10^{-12}\,\mathrm{m^2/s}$ with mobility $\mu=1/\gamma$ — a transport coefficient read straight off the decay rate of an equilibrium fluctuation.

Example 2 — Single-carrier conductivity (Nernst–Einstein). This assembles the Green–Kubo conductivity from the diffusion constant for a dilute charge carrier.

Step 1. Take $n=1.0\times10^{26}\,\mathrm{m^{-3}}$ carriers of charge $q=e=1.602\times10^{-19}\,\mathrm{C}$ , each with the diffusion constant $D=4.14\times10^{-12}\,\mathrm{m^2/s}$ from Example 1, at $T=300\,\mathrm{K}$ .

Step 2. The Green–Kubo conductivity for independent carriers is $\sigma=\dfrac{nq^2}{k_BT}\int_0^\infty\langle v(t)v(0)\rangle\,\mathrm{d}t=\dfrac{nq^2D}{k_BT}=nq^2\mu$ (Nernst–Einstein).

Step 3. Compute $nq^2=(10^{26})(1.602\times10^{-19})^2=(10^{26})(2.566\times10^{-38})=2.566\times10^{-12}$ ; then $\sigma=nq^2 D/k_BT=(2.566\times10^{-12})(4.143\times10^{-12})/(4.143\times10^{-21})$ .

Step 4. Evaluate: $\sigma=(2.566\times10^{-12})(4.143\times10^{-12})/(4.143\times10^{-21})=2.566\times10^{-12}\times10^{9}=2.566\times10^{-3}\,\mathrm{S/m}$ . The dilute carrier gas conducts at $\sigma\approx2.6\times10^{-3}\,\mathrm{S/m}$ — assembled entirely from the equilibrium velocity correlation, with no field ever applied.

:::physicist[kubo formula|derived by Ryogo Kubo in 1957|The Kubo formula is the exact microscopic expression for any linear transport coefficient or susceptibility as a retarded equilibrium average of a commutator — $\chi_{BA}(t)=\tfrac{i}{\hbar}\Theta(t)\langle[B(t),A(0)]\rangle_0$ . Ryogo Kubo derived it in his 1957 paper "Statistical-Mechanical Theory of Irreversible Processes" (J. Phys. Soc. Jpn. 12, 570), unifying the transport formulas of Green, the fluctuation-dissipation theorem of Callen and Welton, and Onsager's reciprocity into a single linear-response identity. Its classical, low-frequency limit is the family of Green–Kubo relations — conductivity, viscosity, thermal conductivity, and diffusion as time integrals of equilibrium current autocorrelations — now the standard route to transport coefficients in molecular-dynamics simulation.]

Origins · 1 minhow this was discovered

History

1905

— Einstein's relation

D=\mu k_BT

prefigures the fluctuation–transport link

1931

— Onsager derives the reciprocal relations for kinetic coefficients

1951

— Callen and Welton prove the fluctuation-dissipation theorem

1952

— Green expresses transport coefficients as time-correlation integrals

1957

— Kubo writes the general linear-response formula (J. Phys. Soc. Jpn. 12, 570)

1966

— Kubo's review consolidates the correlation-function method

Going further

References. R. Kubo, J. Phys. Soc. Jpn. 12, 570 (1957) · R. Kubo, M. Toda, N. Hashitsume, Statistical Physics II: Nonequilibrium Statistical Mechanics (Springer) · Pathria & Beale, Statistical Mechanics, 3rd ed., ch. on linear response · Kardar, Statistical Physics of Particles, linear-response chapter · Reif, Fundamentals of Statistical and Thermal Physics · Landau & Lifshitz, Statistical Physics Part 1, §123–126 · Chaikin & Lubensky, Principles of Condensed Matter Physics, Ch. 7

Provenance. Derivations follow standard linear-response theory as in Kubo, Toda & Hashitsume (Statistical Physics II), Pathria & Beale, Kardar, and Landau–Lifshitz Statistical Physics Part 1 §123–126. A verifier recomputes: (1) the first-order density-matrix solution and the retarded kernel $\chi_{BA}=\tfrac{i}{\hbar}\Theta(t)\langle[B(t),A(0)]\rangle$ ; (2) the Green–Kubo diffusion integral $D=\int_0^\infty\langle v(t)v(0)\rangle\mathrm{d}t=k_BT/(m\gamma)$ from the exponential autocorrelation; (3) the single-carrier conductivity $\sigma=nq^2D/k_BT=nq^2\mu$ (Nernst–Einstein); (4) the FDT prefactor $(1-e^{-\beta\hbar\omega})/2\hbar$ and its classical limit $\beta\omega/2$ . All numbers quoted are standard textbook values.

Builds on: The Fluctuation-Dissipation Theorem · The Langevin Equation · The Einstein Relation · The Ornstein-Uhlenbeck Process. Leads to: Brownian Motion · The Diffusion Equation · The Fokker-Planck Equation.

How can a transport coefficient be an equilibrium average when transport is a nonequilibrium process?

This is the central surprise of the formula, and it is legitimate because the coefficient is a linear-response quantity. Conductivity is defined as the ratio of induced current to applied field in the limit of vanishing field — that is, to first order. And to first order the induced current is computed by tracing against the unperturbed equilibrium density matrix, so the answer is an equilibrium correlation function. The physical content is fluctuation-dissipation: the way spontaneous currents fluctuate and decay in equilibrium already encodes how the system will dissipate energy under a weak drive. You are not computing the driven steady state; you are computing the equilibrium fluctuations that determine the linear coefficient.

What is the difference between the Kubo formula and the Green–Kubo relations?

The Kubo formula is the general, exact, quantum statement: any linear response function is the retarded equilibrium average of a commutator, $\chi_{BA}(t)=\tfrac{i}{\hbar}\Theta(t)\langle[B(t),A(0)]\rangle_0$ . The Green–Kubo relations are its classical, DC-limit specializations for transport coefficients: conductivity, viscosity, thermal conductivity, and diffusion each written as $\int_0^\infty\langle J(0)J(t)\rangle\,\mathrm{d}t$ for the appropriate flux $J$ . So Green–Kubo is what you get by taking $\hbar\to0$ and $\omega\to0$ in the Kubo formula and choosing $A$ and $B$ to be currents. In practice "Green–Kubo" names the recipe used in classical molecular dynamics, while "Kubo formula" names the underlying quantum linear-response identity.

Why does the current autocorrelation have to decay for the conductivity to be finite?

Because the conductivity is the integral of the correlation, $\sigma\propto\int_0^\infty\langle J(t)J(0)\rangle\,\mathrm{d}t$ . If the correlation decays to zero, the integral converges to a finite number — ordinary dissipative transport. If it does not decay (a persistent current memory), the integral diverges, signaling either a superconductor/ballistic conductor with an infinite DC response or an ill-posed limit. This non-decaying piece is the "Drude weight," a $\delta$ -function in $\sigma(\omega)$ at $\omega=0$ , and it is physically distinct from the smooth dissipative background. Taking $V\to\infty$ before $\omega\to0$ is the mathematical device that separates the two: in a truly conducting-but-resistive system the correlation decays and the Green–Kubo integral is finite.

How does Onsager reciprocity emerge from the Kubo framework?

Onsager's relations $L_{ij}=L_{ji}$ (in zero magnetic field) say the matrix of transport coefficients coupling fluxes and forces is symmetric — for example the cross-coefficient that produces a heat current from an electric field equals the one that produces an electric current from a temperature gradient. In the Kubo framework each $L_{ij}$ is a current–current correlation integral $\int\langle J_i(t)J_j(0)\rangle\,\mathrm{d}t$ . Microscopic time-reversal invariance of the equilibrium dynamics makes these correlators symmetric under $i\leftrightarrow j$ (with a sign flip of the magnetic field), so $L_{ij}(B)=L_{ji}(-B)$ follows directly. The Kelvin relation between the Peltier and Seebeck coefficients, $\Pi=TS$ , is the best-known consequence.

Is the Kubo formula only for electrical conductivity?

No — it is completely general in the choice of coupling and observed operators. Pick $A$ to be the polarization and $B$ the current for electrical conductivity; the stress tensor for shear viscosity; the heat current for thermal conductivity; a tagged particle's position/velocity for self-diffusion; the magnetization for magnetic susceptibility; the density for the dynamic structure factor. In each case the transport coefficient or susceptibility is the same Kubo integral over the equilibrium autocorrelation of the appropriate flux. This universality — one formula, many labels — is exactly why linear-response theory reorganized transport theory: conductivity, viscosity, and diffusion stopped being separate subjects and became instances of one equilibrium correlation function.

What does the imaginary-time integral in the full quantum Kubo conductivity mean?

The exact quantum conductivity is often written with an imaginary-time (Kubo–Mori) integral, $\sigma\sim\int_0^\infty\!\mathrm{d}t\,e^{i\omega t}\int_0^\beta\!\mathrm{d}\lambda\,\langle J(0)J(t+i\hbar\lambda)\rangle$ . The $\lambda$ -integral, running over $0$ to $\beta$ , comes from the operator identity $[A,e^{-\beta H_0}]=e^{-\beta H_0}\int_0^\beta e^{\lambda H_0}[H_0,A]e^{-\lambda H_0}\,\mathrm{d}\lambda$ , which rewrites the commutator with the thermal density matrix as a smooth imaginary-time-ordered correlation. It is the "canonical correlation" that makes the quantum expression manifestly Hermitian-friendly. In the classical limit $\hbar\to0$ the imaginary-time interval $[0,\beta]$ collapses to a single factor $\beta=1/k_BT$ , recovering the simple real-time Green–Kubo integral with its $1/k_BT$ prefactor.