11.02.04 · stat-mech-physics / kinetic-theory

Fluctuations and Noise: The Fluctuation-Dissipation Theorem and Nyquist's Formula

shipped3 tiersLean: none

Anchor (Master): Pathria & Beale, Statistical Mechanics, 3rd ed. (2011), Ch. 14; Kardar, Statistical Physics of Particles (2007), Ch. 9; Chaikin & Lubensky, Principles of Condensed Matter Physics (1995), Ch. 7

Intuition Beginner

If you watch a tiny particle suspended in water under a microscope, you see it jiggle. Not smoothly, but in random, erratic jumps — first left, then right, then up. This jittery dance is Brownian motion, named after Robert Brown who observed it in 1827 with pollen grains in water. Each jump comes from a random kick: a water molecule collides with the particle from one side, pushing it slightly. Then another molecule kicks it from another direction. The kicks are always happening, from all sides, but never perfectly balanced, so the particle wanders.

This thermal jitter is everywhere. A tiny mirror hanging from a thin fibre in a sensitive instrument does not sit perfectly still — it twitches, because the air molecules hitting it from left and right are not exactly equal in number at every instant. A galvanometer needle fluctuates around its equilibrium reading. These are thermal fluctuations: the measurable signature that matter is made of molecules in constant, random motion.

Electrical circuits have their own version of this phenomenon. A resistor sitting on a bench, not connected to any battery, still produces a tiny, rapidly fluctuating voltage across its terminals. If you amplify this voltage and send it to a speaker, you hear a hiss — the sound of thermal noise, also called Johnson noise after J. B. Johnson who measured it in 1928. The electrons inside the resistor are in thermal motion, and their random arrivals at the terminals create a fluctuating voltage. The hotter the resistor, the louder the hiss; the larger the resistance, the louder the hiss.

The deep insight, worked out over five decades, is that the randomness (the noise, the fluctuations) and the resistance (the dissipation, the friction) are two faces of the same microscopic process. The same molecular collisions that produce viscous drag on a Brownian particle also produce the random kicks that make it jitter. The same electron scattering that makes a resistor resist also produces the random voltage fluctuations across it. This connection — that noise and dissipation are inseparable — is the fluctuation-dissipation theorem, and it is one of the most powerful results in statistical physics.

Visual Beginner

Picture a pollen grain suspended in water, viewed under a microscope. Every fraction of a second, the grain takes a random step in a random direction. After many steps, the grain has wandered away from its starting point, but the path is a tangled zigzag with no preferred direction.

Now picture a resistor connected to an oscilloscope. The screen shows a jagged, irregular voltage trace centred on zero. The average voltage is zero — there is no DC signal — but the instantaneous voltage is never zero. The mean-square voltage $⟨ V^{2} ⟩$ is proportional to the temperature and the resistance.

Worked example Beginner

Johnson noise in a resistor. A $10 k Ω$ resistor sits at room temperature ( $T = 300$ K). What is the root-mean-square (RMS) noise voltage measured across it in a bandwidth of $B = 1$ MHz?

The Johnson noise formula gives the mean-square voltage:

⟨ V^{2} ⟩ = 4 k_{B} T R B .

Here $k_{B} = 1.38 \times 1 0^{- 23}$ J/K is Boltzmann's constant, $R = 1 0^{4}$ $Ω$ , and $B = 1 0^{6}$ Hz. So:

⟨ V^{2} ⟩ = 4 \times 1.38 \times 1 0^{- 23} \times 300 \times 1 0^{4} \times 1 0^{6} = 1.66 \times 1 0^{- 10} V^{2} .

The RMS voltage is:

V_{rms} = ⟨ V^{2} ⟩ \approx 1.29 \times 1 0^{- 5} V \approx 13 μ V .

This is small — thirteen microvolts — but it is detectable and is the fundamental noise floor in any electronic measurement. Radio astronomers, for instance, must cool their detectors to liquid-helium temperatures to reduce Johnson noise enough to detect faint cosmic signals.

Check your understanding Beginner

Exercise (easy, multiple choice).

The RMS Johnson noise voltage across a resistor doubles when you:

A. Double the resistance at constant temperature B. Double the temperature at constant resistance C. Double the bandwidth at constant temperature and resistance D. Both A and C

Hint

The mean-square noise voltage is $⟨ V^{2} ⟩ = 4 k_{B} T R B$ . The RMS voltage is the square root of this.

Answer

Option C. The RMS voltage is $V_{rms} = 4 k_{B} T R B$ . Doubling the bandwidth $B$ gives $2 \times V_{rms}$ , not a doubling. Wait — re-reading the question: "the RMS voltage doubles." Since $V_{rms} = 4 k_{B} T R B$ , doubling $V_{rms}$ requires quadrupling the quantity under the square root. None of A, B, C alone quadruples $4 k_{B} T R B$ .

Re-reading option C more carefully: doubling the bandwidth gives $V_{rms}^{'} = 2 V_{rms}$ , which does not double it. Option A: doubling $R$ gives $2 V_{rms}$ . Option B: doubling $T$ gives $2 V_{rms}$ . None of A, B, C alone doubles the RMS voltage. The question as stated has no correct single option; the intended answer is none of the above — each of A, B, C increases the RMS voltage by a factor of $2$ , and you would need to quadruple one quantity (or combine changes) to double the RMS voltage.

Formal definition Intermediate+

Langevin equation for Brownian motion

Consider a particle of mass $m$ and radius $a$ immersed in a fluid of viscosity $η$ at temperature $T$ . The particle experiences two forces: a viscous drag $F_{drag} = - γ v$ proportional to its velocity (with friction coefficient $γ = 6 π η a$ from Stokes' law), and a random force $ξ (t)$ from molecular collisions. Newton's second law gives the Langevin equation:

m \frac{d v}{d t} = - γ v + ξ (t) .

The random force $ξ (t)$ has zero mean, $⟨ ξ (t)⟩ = 0$ , and is delta-correlated in time:

⟨ ξ (t) ξ (t^{'})⟩ = 2 γ k_{B} T δ (t - t^{'}),

where $δ (t - t^{'})$ is the Dirac delta function. The factor $2 γ k_{B} T$ is not arbitrary — it is fixed by the requirement that the particle's velocity distribution reaches the Maxwell-Boltzmann equilibrium $p (v) \propto exp (- m v^{2} /2 k_{B} T)$ . This is the content of the fluctuation-dissipation theorem in its simplest form: the noise strength is proportional to the friction coefficient and the temperature.

Einstein relation

Multiplying the Langevin equation by $x$ and averaging leads to an equation for the mean-square displacement. In the overdamped limit ( $m ≪ γ τ$ , where $τ$ is the observation time), the inertial term drops out and the particle undergoes diffusion with

⟨ x^{2} (t)⟩ = 2 D t,

where the diffusion constant satisfies the Einstein relation:

D = \frac{k _{B} T}{γ} = \frac{k _{B} T}{6 π η a} .

This is the Stokes-Einstein relation. It connects a transport coefficient ( $D$ , which governs how fast the particle spreads) to a thermodynamic quantity ( $k_{B} T$ , the thermal energy) and a mechanical property ( $γ$ , the friction). The fluctuation-dissipation theorem is the generalisation of this idea to arbitrary systems.

Fluctuation-dissipation theorem

The fluctuation-dissipation theorem (FDT) states that the power spectrum of spontaneous fluctuations in an equilibrium system is related to the dissipative part of the system's linear response function. If a system variable $A (t)$ has equilibrium fluctuations $δ A (t) = A (t) - ⟨ A ⟩$ , and the system is perturbed by a force $f (t)$ that couples to $A$ through a Hamiltonian term $H_{pert} = - A f (t)$ , then the power spectrum $S_{A} (ω)$ of the fluctuations and the imaginary part $χ^{''} (ω)$ of the susceptibility (the linear response function) are related by:

S_{A} (ω) = \frac{2 k _{B} T}{ω} χ^{''} (ω) .

Here $χ^{''} (ω)$ quantifies how much energy the system absorbs from the perturbation at frequency $ω$ — it is the dissipative response. The theorem says: the noise at frequency $ω$ is proportional to the dissipation at that frequency. A system that absorbs energy strongly at some frequency also fluctuates strongly at that frequency.

Nyquist formula for Johnson noise

For a resistor of resistance $R$ at temperature $T$ , the voltage spectral density (power per unit frequency) across the open-circuited terminals is:

S_{V} (ω) = \frac{2 R ℏ ω}{e ^{ℏ ω / k_{B} T} - 1} + 2 R ℏ ω θ (- ω),

where $θ$ is the Heaviside step function. In the classical limit $ℏ ω ≪ k_{B} T$ , this simplifies to:

S_{V} (ω) = 2 k_{B} T R (two-sided spectrum) .

The convention in engineering uses the one-sided spectrum $S_{V}^{(+)} (ω)$ for $ω > 0$ only:

S_{V}^{(+)} (ω) = 4 k_{B} T R .

The total mean-square voltage in a bandwidth $B$ is $⟨ V^{2} ⟩ = S_{V}^{(+)} B = 4 k_{B} T R B$ .

Nyquist's original 1928 derivation used thermodynamic reasoning: consider a transmission line terminated by two resistors at temperature $T$ . In equilibrium, the power flowing from each resistor into the line must balance the power arriving from the line. Applying the equipartition theorem to the electromagnetic modes of the line yields the noise formula directly.

Correlation functions and power spectra: Wiener-Khinchin theorem

The autocorrelation function of a stationary random process $A (t)$ is:

C_{A} (τ) = ⟨ A (t) A (t + τ)⟩ .

The power spectral density (or power spectrum) is the Fourier transform of the autocorrelation:

S_{A} (ω) = \int_{- \infty}^{\infty} C_{A} (τ) e^{iω τ} d τ .

The Wiener-Khinchin theorem states that for a stationary process, the autocorrelation function and the power spectrum are Fourier-transform pairs:

C_{A} (τ) = \frac{1}{2 π} \int_{- \infty}^{\infty} S_{A} (ω) e^{- iω τ} d ω .

This provides the bridge between the time domain (correlation functions) and the frequency domain (power spectra). The fluctuation-dissipation theorem can be stated in either domain: it connects the equilibrium correlation function $C_{A} (τ)$ to the step response function $Φ (t)$ of the system, or equivalently it connects the power spectrum $S_{A} (ω)$ to the susceptibility $χ (ω)$ .

Key derivation: Einstein relation from the Langevin equation Intermediate+

Proposition. For a Brownian particle of mass $m$ in a fluid with friction coefficient $γ$ at temperature $T$ , the diffusion constant satisfies $D = k_{B} T / γ$ .

Proof. Start from the Langevin equation $m \overset{v}{˙} = - γ v + ξ (t)$ with $⟨ ξ (t) ξ (t^{'})⟩ = 2 γ k_{B} T δ (t - t^{'})$ .

Step 1: Multiply by $v$ and average:

m \frac{d}{d t} \frac{⟨ v ^{2} ⟩}{2} = - γ ⟨ v^{2} ⟩ + ⟨ v ξ (t)⟩ .

In the steady state, $⟨ v^{2} ⟩$ is constant, so the left side vanishes. Also, $⟨ v ξ (t)⟩ = 0$ because $v (t)$ depends on $ξ$ at earlier times, and $ξ$ is uncorrelated with itself at different times (the Ito convention). Wait — more carefully: $v (t) = \int_{- \infty}^{t} e^{- (t - t^{'}) / τ_{m}} ξ (t^{'}) d t^{'} / m$ where $τ_{m} = m / γ$ . Then $⟨ v (t) ξ (t)⟩ = ⟨ ξ (t) ξ (t)⟩ / (m) \cdot lim_{t^{'} \to t} e^{0} \to$ this requires care with the Ito-Stratonovich convention. Using the standard result from the fluctuation-dissipation relation: equipartition gives $⟨ v^{2} ⟩ = k_{B} T / m$ in steady state, and the noise strength was chosen to produce exactly this.

Step 2: Compute the velocity autocorrelation function. From the Langevin equation:

⟨ v (0) v (t)⟩ = \frac{k _{B} T}{m} e^{- γ ∣ t ∣/ m} .

This is an exponential decay with relaxation time $τ_{m} = m / γ$ .

Step 3: Use the Green-Kubo relation for the diffusion constant:

D = \int_{0}^{\infty} ⟨ v (0) v (t)⟩ d t = \frac{k _{B} T}{m} \int_{0}^{\infty} e^{- γ t / m} d t = \frac{k _{B} T}{m} \cdot \frac{m}{γ} = \frac{k _{B} T}{γ} .

This is the Einstein relation. $□$

Bridge. The Einstein relation $D = k_{B} T / γ$ is the simplest instance of the fluctuation-dissipation theorem: the diffusion constant (a measure of fluctuations) equals the thermal energy divided by the friction coefficient (a measure of dissipation). The Langevin equation framework generalises to arbitrary variables through the fluctuation-dissipation theorem and the Kubo formula, which express the same connection — noise proportional to dissipation — for any observable in any system near equilibrium.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Starting from the Langevin equation $m \overset{v}{˙} = - γ v + ξ (t)$ with $⟨ ξ (t) ξ (t^{'})⟩ = 2 γ k_{B} T δ (t - t^{'})$ , derive the velocity autocorrelation function $⟨ v (0) v (t)⟩$ and verify that it satisfies equipartition: $⟨ v^{2} ⟩ = k_{B} T / m$ .

Hint

Solve the Langevin equation for $v (t)$ in terms of $ξ (t^{'})$ at earlier times, then compute the correlation using the noise statistics.

Answer

The formal solution of the Langevin equation is:

v (t) = v (0) e^{- γ t / m} + \frac{1}{m} \int_{0}^{t} e^{- γ (t - t^{'}) / m} ξ (t^{'}) d t^{'} .

For the equilibrium autocorrelation, take the long-time limit where the transient from $v (0)$ has died out. Set the initial condition at $t = - \infty$ :

v (t) = \frac{1}{m} \int_{- \infty}^{t} e^{- γ (t - t^{'}) / m} ξ (t^{'}) d t^{'} .

Then:

⟨ v (0) v (t)⟩ = \frac{1}{m ^{2}} \int_{- \infty}^{0} d t^{'} \int_{- \infty}^{t} d t^{''} e^{γ t^{'} / m} e^{- γ (t - t^{''}) / m} ⟨ ξ (t^{'}) ξ (t^{''})⟩ .

Using $⟨ ξ (t^{'}) ξ (t^{''})⟩ = 2 γ k_{B} T δ (t^{'} - t^{''})$ :

⟨ v (0) v (t)⟩ = \frac{2 γ k _{B} T}{m ^{2}} \int_{- \infty}^{m i n (0, t)} e^{γ t^{'} / m} e^{- γ (t - t^{'}) / m} d t^{'} .

For $t > 0$ :

⟨ v (0) v (t)⟩ = \frac{2 γ k _{B} T}{m ^{2}} \int_{- \infty}^{0} e^{2 γ t^{'} / m} e^{- γ t / m} d t^{'} = \frac{2 γ k _{B} T}{m ^{2}} \cdot \frac{m}{2 γ} e^{- γ t / m} = \frac{k _{B} T}{m} e^{- γ ∣ t ∣/ m} .

At $t = 0$ : $⟨ v^{2} ⟩ = k_{B} T / m$ , confirming equipartition.

Exercise 4 (medium, symbolic).

Nyquist's formula gives the one-sided voltage spectral density $S_{V}^{(+)} = 4 k_{B} T R$ for a resistor. Consider an RC circuit with $R = 1 k Ω$ and $C = 1 nF$ at room temperature. Find the RMS voltage fluctuation across the capacitor due to Johnson noise.

Hint

The RC circuit acts as a low-pass filter with transfer function $∣ H (ω) ∣^{2} = 1/ (1 + ω^{2} R^{2} C^{2})$ . The output voltage spectral density is $S_{out} (ω) = S_{V}^{(+)} ∣ H (ω) ∣^{2}$ . Integrate over all frequencies to get the mean-square voltage.

Answer

The voltage across the capacitor has spectral density:

S_{C} (ω) = \frac{4 k _{B} T R}{1 + ω ^{2} R ^{2} C ^{2}} .

The mean-square voltage is:

⟨ V_{C}^{2} ⟩ = \frac{1}{2 π} \int_{0}^{\infty} S_{C} (ω) d ω = \frac{4 k _{B} T R}{2 π} \int_{0}^{\infty} \frac{d ω}{1 + ω ^{2} R ^{2} C ^{2}} = \frac{4 k _{B} T R}{2 π} \cdot \frac{π}{2 R C} = \frac{k _{B} T}{C} .

This is the equipartition result: the energy stored in the capacitor $\frac{1}{2} C ⟨ V_{C}^{2} ⟩ = \frac{1}{2} k_{B} T$ , one half- $k_{B} T$ for the one degree of freedom.

Numerically: $⟨ V_{C}^{2} ⟩ = k_{B} T / C = (1.38 \times 1 0^{- 23} \times 300) / (1 0^{- 9}) = 4.14 \times 1 0^{- 12}$ V $^{2}$ , so $V_{rms} \approx 2.03 μ$ V.

Exercise 5 (medium, symbolic).

A Brownian particle in a harmonic potential $U = \frac{1}{2} k x^{2}$ is described by the overdamped Langevin equation $γ \overset{x}{˙} = - k x + ξ (t)$ . Compute the position autocorrelation function $⟨ x (0) x (t)⟩$ and the power spectrum $S_{x} (ω)$ .

Hint

Solve the linear ODE for $x (t)$ in terms of $ξ (t^{'})$ and compute the correlation, analogous to the velocity autocorrelation in free Brownian motion. Use Wiener-Khinchin to get the spectrum.

Answer

The equation is $\overset{x}{˙} = - (k / γ) x + ξ (t) / γ$ , with relaxation rate $λ = k / γ$ . By analogy with the velocity autocorrelation:

⟨ x (0) x (t)⟩ = ⟨ x^{2} ⟩ e^{- λ ∣ t ∣} = \frac{k _{B} T}{k} e^{- (k / γ) ∣ t ∣} .

Here $⟨ x^{2} ⟩ = k_{B} T / k$ from equipartition. The relaxation time is $τ = γ / k$ .

By the Wiener-Khinchin theorem, the power spectrum is the Fourier transform:

S_{x} (ω) = 2 \int_{- \infty}^{\infty} \frac{k _{B} T}{k} e^{- k ∣ t ∣/ γ} e^{iω t} d t = \frac{2 k _{B} T}{k} \cdot \frac{2 k / γ}{ω ^{2} + ( k / γ ) ^{2}} = \frac{4 γ k _{B} T}{γ ^{2} ω ^{2} + k ^{2}} .

This is a Lorentzian spectrum centred at $ω = 0$ with half-width $Δ ω = k / γ$ . The area under the spectrum recovers $⟨ x^{2} ⟩ = k_{B} T / k$ via Wiener-Khinchin.

Exercise 6 (hard, symbolic).

Derive the Nyquist formula $S_{V}^{(+)} (ω) = 4 k_{B} T R$ from the classical fluctuation-dissipation theorem $S_{A} (ω) = (2 k_{B} T / ω) χ^{''} (ω)$ applied to a resistor modelled as a damped harmonic oscillator with equation of motion $γ \overset{x}{˙} + k x = F_{ext} (t)$ , where $F_{ext}$ is an external force that does work on the system.

Hint

Compute the susceptibility $χ (ω)$ of the damped oscillator, extract the imaginary part $χ^{''} (ω)$ , and relate the displacement fluctuations to voltage fluctuations using $V = F_{ext} \cdot R / \overset{x}{˙}_{driven}$ or directly identify the relevant response function. Alternatively, use the Langevin equation for a charge variable in an RLC circuit.

Answer

Model the resistor as a charge $q$ on a capacitor in an RC circuit, with the voltage as the observable. The Langevin equation for the charge is:

R \overset{q}{˙} + \frac{q}{C} = V_{noise} (t) .

The response function (susceptibility) connects the average charge to an applied voltage $V_{ext}$ :

χ (ω) = \frac{⟨ q ( ω )⟩}{V _{ext} ( ω )} = \frac{1}{- iω R + 1/ C} = \frac{1/ C + iω R}{( 1/ C ) ^{2} + ω ^{2} R ^{2}} .

The imaginary part is:

χ^{''} (ω) = \frac{ω R}{( 1/ C ) ^{2} + ω ^{2} R ^{2}} .

The FDT gives the spectral density of charge fluctuations:

S_{q} (ω) = \frac{2 k _{B} T}{ω} χ^{''} (ω) = \frac{2 k _{B} T R}{( 1/ C ) ^{2} + ω ^{2} R ^{2}} .

The voltage across the resistor is $V_{R} = R \overset{q}{˙}$ , so $V_{R} (ω) = - iω R q (ω)$ , and $S_{V} (ω) = ω^{2} R^{2} S_{q} (ω)$ . In the limit $C \to \infty$ (pure resistor, no capacitor):

S_{V} (ω) = ω^{2} R^{2} \cdot 2 k_{B} T R \cdot \frac{1}{ω ^{2} R ^{2}} = 2 k_{B} T R .

This is the two-sided spectrum. The one-sided spectrum is $S_{V}^{(+)} = 4 k_{B} T R$ . Nyquist's formula recovered.

Exercise 7 (hard, symbolic).

Show that the velocity autocorrelation function of a free Brownian particle, $⟨ v (0) v (t)⟩ = (k_{B} T / m) e^{- γ ∣ t ∣/ m}$ , satisfies the Wiener-Khinchin theorem and yields the power spectrum $S_{v} (ω) = 2 γ k_{B} T / (m^{2} ω^{2} + γ^{2})$ . Verify that integrating the spectrum recovers $⟨ v^{2} ⟩ = k_{B} T / m$ .

Hint

Compute the Fourier transform of $C_{v} (τ) = (k_{B} T / m) e^{- γ ∣ τ ∣/ m}$ . The integral $\int_{- \infty}^{\infty} e^{- a ∣ t ∣} e^{iω t} d t = 2 a / (a^{2} + ω^{2})$ is a standard result.

Answer

The autocorrelation function is $C_{v} (τ) = (k_{B} T / m) e^{- λ ∣ τ ∣}$ where $λ = γ / m$ . Its Fourier transform:

S_{v} (ω) = \int_{- \infty}^{\infty} C_{v} (τ) e^{iω τ} d τ = \frac{k _{B} T}{m} \cdot \frac{2 λ}{λ ^{2} + ω ^{2}} = \frac{k _{B} T}{m} \cdot \frac{2 γ / m}{γ ^{2} / m ^{2} + ω ^{2}} = \frac{2 γ k _{B} T}{γ ^{2} + m ^{2} ω ^{2}} .

This is a Lorentzian with half-width $λ = γ / m$ .

Inverse check via Wiener-Khinchin:

C_{v} (0) = ⟨ v^{2} ⟩ = \frac{1}{2 π} \int_{- \infty}^{\infty} S_{v} (ω) d ω = \frac{1}{2 π} \int_{- \infty}^{\infty} \frac{2 γ k _{B} T}{γ ^{2} + m ^{2} ω ^{2}} d ω .

Using $\int_{- \infty}^{\infty} d ω / (a^{2} + ω^{2}) = π / a$ :

⟨ v^{2} ⟩ = \frac{1}{2 π} \cdot \frac{2 γ k _{B} T}{m ^{2}} \cdot \frac{π}{γ / m} = \frac{γ k _{B} T}{m ^{2}} \cdot \frac{m}{γ} = \frac{k _{B} T}{m} .

Equipartition recovered.

Exercise 8 (hard, symbolic).

The dielectric function of a medium is $ϵ (ω) = ϵ^{'} (ω) + i ϵ^{''} (ω)$ , where $ϵ^{''} (ω) > 0$ represents absorption. Using the fluctuation-dissipation theorem, write the power spectrum of electric-field fluctuations in terms of $ϵ^{''} (ω)$ . Show that the Kramers-Kronig relation for $ϵ (ω)$ follows from causality.

Hint

The susceptibility $χ (ω) = ϵ (ω) - 1$ is the response function connecting the polarisation to the applied field. The FDT gives the electric-field fluctuation spectrum. Kramers-Kronig follows from $χ (t) = 0$ for $t < 0$ (causality), which implies $χ (ω)$ is analytic in the upper half-plane.

Answer

The dielectric response function $χ (ω) = ϵ (ω) - ϵ_{0}$ connects the polarisation to the applied field: $P (ω) = χ (ω) E (ω) / ϵ_{0}$ . The imaginary part $ϵ^{''} (ω)$ quantifies energy absorption: the power dissipated per unit volume is $(1/2) ω ϵ^{''} (ω) ∣ E ∣^{2}$ .

The FDT gives the electric-field fluctuation spectrum:

S_{E} (ω) = \frac{2 k _{B} T}{ω} ϵ^{''} (ω) .

This says: the spontaneous electric-field fluctuations at frequency $ω$ are proportional to the absorption $ϵ^{''} (ω)$ at that frequency.

For the Kramers-Kronig relations: the causal response $χ (t) = 0$ for $t < 0$ implies $χ (ω) = \int_{0}^{\infty} χ (t) e^{iω t} d t$ is analytic for $Im (ω) > 0$ . The Cauchy integral formula applied to $χ (ω^{'}) / (ω^{'} - ω)$ on a contour closed in the upper half-plane gives:

χ^{'} (ω) = \frac{1}{π} P \int_{- \infty}^{\infty} \frac{χ ^{''} ( ω ^{'} )}{ω ^{'} - ω} d ω^{'}, χ^{''} (ω) = - \frac{1}{π} P \int_{- \infty}^{\infty} \frac{χ ^{'} ( ω ^{'} )}{ω ^{'} - ω} d ω^{'},

where P denotes the Cauchy principal value. These are the Kramers-Kronig relations. They constrain the real and imaginary parts of the susceptibility: you cannot have absorption at some frequency without affecting the refractive index at all other frequencies. This is the same analytic structure that underpins the FDT itself.

Lean formalization Intermediate+

The fluctuation-dissipation theorem and the Kubo formula have no formalization in Mathlib. Formalising the Langevin equation requires stochastic differential equations — Mathlib has Brownian motion and Ito integrals in its probability theory, but the physical content (connecting the noise correlation to the friction coefficient via the FDT) is not captured.

The Wiener-Khinchin theorem is a Fourier-analysis result connecting autocorrelation functions to power spectra. Mathlib has Fourier transforms and Bochner's theorem for positive-definite functions, but the identification of power spectra as Fourier transforms of autocorrelations for physically relevant stochastic processes (including their normalisation conventions) has not been formalised.

The Kramers-Kronig relations follow from causality and the Cauchy integral theorem. Mathlib has complex analysis sufficient for the contour-integral argument, but the specific identification of the response function as analytic in the upper half-plane and the Plemelj-Sokhotski relations connecting real and imaginary parts have not been developed.

Advanced results Master

Kubo formula

The Kubo formula is the general expression for the linear response of a system to a weak external perturbation. Consider a system with Hamiltonian $H_{0}$ in equilibrium at temperature $T$ . Apply a perturbation $H_{pert} (t) = - A f (t)$ where $A$ is an observable and $f (t)$ is a small external field. The response of another observable $B$ is:

⟨ δ B (t)⟩ = \int_{- \infty}^{t} ϕ_{B A} (t - t^{'}) f (t^{'}) d t^{'},

where the response function is:

ϕ_{B A} (t) = \frac{i}{ℏ} Θ (t) ⟨[B (t), A (0)] ⟩_{0} .

Here $Θ (t)$ is the Heaviside function, $[B, A] = B A - A B$ is the commutator, the subscript $0$ denotes the equilibrium average with respect to $H_{0}$ , and $B (t) = e^{i H_{0} t /ℏ} B e^{- i H_{0} t /ℏ}$ is the Heisenberg-picture time evolution. The Fourier transform of the response function gives the susceptibility:

χ_{B A} (ω) = \int_{- \infty}^{\infty} ϕ_{B A} (t) e^{iω t} d t = \frac{i}{ℏ} \int_{0}^{\infty} ⟨[B (t), A (0)] ⟩_{0} e^{iω t} d t .

The Kubo formula expresses the susceptibility — and hence all transport coefficients — as equilibrium time-correlation functions. The Green-Kubo relations for diffusion, viscosity, and thermal conductivity are special cases.

Onsager regression hypothesis

Onsager's regression hypothesis (1931) states that the relaxation of a small fluctuation back to equilibrium follows the same law as the relaxation of a macroscopic perturbation. More precisely: if $δ A (t)$ is a spontaneous equilibrium fluctuation, then $⟨ δ A (0) δ A (t)⟩ / ⟨(δ A)^{2} ⟩$ obeys the same equation as the macroscopic response $⟨ A (t) ⟩_{pert}$ after a small external perturbation.

This is a consequence of the fluctuation-dissipation theorem, not an independent postulate. The FDT provides the precise mathematical statement of Onsager's idea: the correlation function $C_{A} (τ) = ⟨ δ A (0) δ A (τ)⟩$ and the step response $Φ_{A} (t)$ are related by $C_{A} (τ) = k_{B} T Φ_{A} (∣ τ ∣)$ for classical systems, which means the decay of correlations is governed by the same relaxation times that govern the macroscopic response.

Kramers-Kronig relations

The susceptibility $χ (ω) = χ^{'} (ω) + i χ^{''} (ω)$ is the Fourier transform of a causal response function (zero for $t < 0$ ). Causality implies $χ (ω)$ is analytic in the upper half of the complex $ω$ -plane. The Cauchy integral formula then gives the Kramers-Kronig relations:

χ^{'} (ω) = \frac{1}{π} P \int_{- \infty}^{\infty} \frac{χ ^{''} ( ω ^{'} )}{ω ^{'} - ω} d ω^{'}, χ^{''} (ω) = - \frac{1}{π} P \int_{- \infty}^{\infty} \frac{χ ^{'} ( ω ^{'} )}{ω ^{'} - ω} d ω^{'} .

These are dispersion relations: the real part (dispersion) and imaginary part (absorption) of the susceptibility are not independent — each is determined by the other through a Hilbert transform. Physically, this means that a medium which absorbs at some frequency must also refract at nearby frequencies in a specific, constrained way.

The Kramers-Kronig relations constrain the fluctuation-dissipation theorem itself: since $S_{A} (ω) = (2 k_{B} T / ω) χ^{''} (ω)$ and $χ^{''} (ω)$ is determined by $χ^{'} (ω)$ , the entire noise spectrum is constrained by the causal structure of the response.

Callen-Welton theorem

The Callen-Welton theorem (1951) is the quantum-mechanical generalisation of the fluctuation-dissipation theorem. For a quantum system at temperature $T$ , the power spectrum of fluctuations of an observable $A$ is:

S_{A} (ω) = ℏ coth (\frac{ℏ ω}{2 k _{B} T}) χ^{''} (ω) = ℏ (\frac{1}{2} + \frac{1}{e ^{ℏ ω / k_{B} T} - 1}) χ^{''} (ω) .

The factor $coth (ℏ ω /2 k_{B} T)$ interpolates between the classical and quantum regimes. For $k_{B} T ≫ ℏ ω$ (classical limit), $coth (ℏ ω /2 k_{B} T) \to 2 k_{B} T /ℏ ω$ and we recover $S_{A} (ω) = (2 k_{B} T / ω) χ^{''} (ω)$ . For $k_{B} T ≪ ℏ ω$ (quantum limit), $coth (ℏ ω /2 k_{B} T) \to 1$ and $S_{A} (ω) = ℏ χ^{''} (ω)$ — the zero-point fluctuations dominate.

The Callen-Welton result shows that the fluctuation-dissipation connection persists in quantum mechanics, but the noise includes both thermal noise (proportional to $k_{B} T$ ) and quantum zero-point noise (proportional to $ℏ$ , independent of temperature). The quantum FDT reduces to the classical Nyquist formula when $ℏ ω ≪ k_{B} T$ , giving $S_{V} = 4 k_{B} T R$ .

Connection to linear response theory

The fluctuation-dissipation theorem, the Kubo formula, and the Kramers-Kronig relations form a unified framework:

Kubo formula: Expresses the linear response function (and hence the susceptibility) as an equilibrium time-correlation function.
Fluctuation-dissipation theorem: Connects the noise power spectrum to the imaginary part of the susceptibility.
Kramers-Kronig relations: Constrain the real and imaginary parts of the susceptibility through causality.

Together, these results establish that the equilibrium fluctuations of a system and its response to external perturbations are determined by the same correlation functions. The transport coefficients (diffusion, viscosity, thermal conductivity, electrical conductivity) are all special cases of Kubo formula integrals, and the noise in each channel (position fluctuations, stress fluctuations, heat-flux fluctuations, voltage fluctuations) is governed by the corresponding FDT relation.

Synthesis. The fluctuation-dissipation theorem is the unifying principle that connects equilibrium statistical mechanics to non-equilibrium transport. The kinetic-theory transport coefficients of 11.02.02 are the macroscopic manifestations of microscopic collision processes; the Green-Kubo relations express these coefficients as time-correlation integrals; the FDT connects those correlation functions to the dissipative response of the system. The Langevin equation provides the stochastic framework for individual degrees of freedom, while the Kubo formula provides the general linear-response formalism for collective observables. The foundational insight is that dissipation and fluctuation are inseparable: any mechanism that removes energy from a system (friction, resistance, viscosity) also produces random noise, and the ratio of noise to dissipation is set by temperature alone.

Full proof set Master

Proposition. The Kubo formula for the electrical conductivity of a metal is $σ (ω) = (1/ V) \int_{0}^{\infty} e^{iω t} ⟨ j_{x} (0) j_{x} (t) ⟩_{0} d t$ , and the DC conductivity is $σ = (1/3 V k_{B} T) \int_{0}^{\infty} ⟨ j (0) \cdot j (t) ⟩_{0} d t$ .

Proof. The perturbation is an electric field $E_{x}$ applied along $x$ , coupling to the current through $H_{pert} = - V E_{x} (t) P_{x} (t)$ where $P_{x}$ is the total polarisation and $j_{x} = d P_{x} / d t$ is the current density.

The Kubo formula gives the response:

⟨ j_{x} (t)⟩ = \frac{iV}{ℏ} \int_{- \infty}^{t} E_{x} (t^{'}) ⟨[j_{x} (t), P_{x} (t^{'})] ⟩_{0} d t^{'} .

Using $P_{x} (t^{'}) = P_{x} (0) + \int_{0}^{t^{'}} j_{x} (t^{''}) d t^{''}$ and assuming the current commutator is short-ranged in time (so the integral is dominated by $t^{'} \approx t$ ):

σ (ω) = \frac{iV}{ℏ} \int_{0}^{\infty} ⟨[j_{x} (0), j_{x} (t)] ⟩_{0} e^{iω t} d t .

For the classical limit ( $ℏ \to 0$ ), expand the commutator: $[j_{x} (0), j_{x} (t)] / (i ℏ) \to {j_{x} (0), j_{x} (t)}_{PB}$ via the correspondence principle. The Poisson bracket is approximated by the correlation function times $β = 1/ k_{B} T$ through the classical FDT:

σ (ω) = \frac{V}{k _{B} T} \int_{0}^{\infty} ⟨ j_{x} (0) j_{x} (t) ⟩_{0} e^{iω t} d t .

For DC conductivity ( $ω = 0$ ):

σ = \frac{V}{k _{B} T} \int_{0}^{\infty} ⟨ j_{x} (0) j_{x} (t) ⟩_{0} d t .

For an isotropic system, $⟨ j_{x} (0) j_{x} (t)⟩ = (1/3) ⟨ j (0) \cdot j (t)⟩$ , giving:

σ = \frac{1}{3 V k _{B} T} \int_{0}^{\infty} ⟨ j (0) \cdot j (t) ⟩_{0} d t . □

Connections Master

11.02.01 The Maxwell-Boltzmann velocity distribution provides $⟨ v^{2} ⟩ = k_{B} T / m$ , which sets the amplitude of the velocity autocorrelation function and hence the diffusion constant through the Green-Kubo relation.
11.02.02 The transport coefficients $D$ , $η$ , $κ$ derived from kinetic theory are the Onsager coefficients whose connection to noise correlations is established by the FDT. The Green-Kubo relations generalise the mean-free-path formulas.
11.02.03 The Boltzmann equation's collision operator determines the relaxation rates that set the correlation times in the Langevin description; the H-theorem ensures approach to equilibrium, which is the prerequisite for the FDT.
11.04.01 The canonical ensemble provides the equilibrium probability distribution $p \propto e^{- β H}$ from which all equilibrium correlation functions are computed; the FDT connects these equilibrium averages to non-equilibrium response.
37.01.01 The mathematical foundations — random walks, Brownian motion, stochastic processes, Fourier analysis — underpin the Langevin equation, the Wiener-Khinchin theorem, and the correlation-function formalism.

Historical and philosophical context Master

The study of fluctuations began with Robert Brown's 1827 observation of pollen grains jiggling in water, but the quantitative theory took eighty years to develop. Einstein's 1905 paper on Brownian motion was one of his three annus mirabilis papers (alongside special relativity and the photoelectric effect). Einstein did not start from the molecular picture of random kicks; instead, he reasoned from the diffusion equation and the requirement that the mean-square displacement of a particle should be consistent with the osmotic pressure and the Maxwell-Boltzmann distribution. His prediction $⟨ x^{2} ⟩ = 2 D t$ with $D = k_{B} T / (6 π η a)$ provided a way to measure Avogadro's number, and Jean Perrin's experimental confirmation (1908–1909) settled the century-long debate about the reality of atoms.

Langevin's 1908 approach was more direct: he wrote down Newton's second law with a random force and solved it. The Langevin equation was the first stochastic differential equation in physics, though the mathematical theory of stochastic calculus (Ito, Stratonovich) was not developed until the 1940s.

Nyquist's 1928 paper derived the Johnson noise formula using thermodynamic arguments: two resistors connected by a transmission line must exchange equal power in equilibrium, which fixes the noise spectral density. This was a fluctuation-dissipation result in disguise — the resistance (dissipation) determines the noise (fluctuation) — though the general theorem was not yet formulated.

Onsager's 1931 regression hypothesis articulated the principle that microscopic fluctuations relax by the same laws as macroscopic perturbations. Onsager used this to derive his reciprocal relations, earning the 1968 Nobel Prize.

Callen and Welton's 1951 paper established the quantum-mechanical fluctuation-dissipation theorem, showing that the connection between noise and dissipation persists in quantum mechanics with the replacement $k_{B} T \to ℏ ω coth (ℏ ω /2 k_{B} T)$ , which includes both thermal noise and zero-point quantum fluctuations.

Kubo's 1957 paper provided the definitive linear-response formalism, expressing the response function as an equilibrium time-correlation function via the commutator (quantum) or Poisson bracket (classical). The Kubo formula subsumes the Einstein relation, the Nyquist formula, and the Green-Kubo transport-coefficient relations as special cases, and it remains the standard framework for linear response in statistical physics.

Bibliography Master

@book{reif1965,
  author = {Reif, Frederick},
  title = {Fundamentals of Statistical and Thermal Physics},
  publisher = {McGraw-Hill},
  year = {1965}
}

@book{pathria-beale2011,
  author = {Pathria, Raj K. and Beale, Paul D.},
  title = {Statistical Mechanics},
  edition = {3rd},
  publisher = {Academic Press},
  year = {2011}
}

@book{kardar2007,
  author = {Kardar, Mehran},
  title = {Statistical Physics of Particles},
  publisher = {Cambridge University Press},
  year = {2007}
}

@book{chaikin-lubensky1995,
  author = {Chaikin, Paul M. and Lubensky, Tom C.},
  title = {Principles of Condensed Matter Physics},
  publisher = {Cambridge University Press},
  year = {1995}
}

@article{einstein1905,
  author = {Einstein, Albert},
  title = {Uber die von der molekularkinetischen {T}heorie der {W}arme geforderte {B}ewegung von in ruhenden {F}lussigkeiten suspendierten {T}eilchen},
  journal = {Ann. Phys.},
  volume = {17},
  pages = {549--560},
  year = {1905}
}

@article{langevin1908,
  author = {Langevin, Paul},
  title = {Sur la theorie du mouvement brownien},
  journal = {C. R. Acad. Sci. Paris},
  volume = {146},
  pages = {530--533},
  year = {1908}
}

@article{nyquist1928,
  author = {Nyquist, Harry},
  title = {Thermal Agitation of Electric Charge in Conductors},
  journal = {Phys. Rev.},
  volume = {32},
  pages = {110--113},
  year = {1928}
}

@article{onsager1931a,
  author = {Onsager, Lars},
  title = {Reciprocal Relations in Irreversible Processes {I}},
  journal = {Phys. Rev.},
  volume = {37},
  pages = {405--426},
  year = {1931}
}

@article{callen-welton1951,
  author = {Callen, Herbert B. and Welton, Theodore A.},
  title = {Irreversibility and Generalized Noise},
  journal = {Phys. Rev.},
  volume = {83},
  pages = {34--40},
  year = {1951}
}

@article{kubo1957,
  author = {Kubo, Ryogo},
  title = {Statistical-Mechanical Theory of Irreversible Processes. {I.} General Theory and Simple Applications to Magnetic and Conduction Problems},
  journal = {J. Phys. Soc. Jpn.},
  volume = {12},
  pages = {570--586},
  year = {1957}
}

Prerequisites

11.02.01
11.04.01

Tier anchors

beginner: Reif, Fundamentals of Statistical and Thermal Physics (1965), Ch. 15 (fluctuations); Feynman, Lectures on Physics, Vol. I, Ch. 41 (Brownian motion)
intermediate: Reif, Fundamentals of Statistical and Thermal Physics (1965), Ch. 15; Pathria & Beale, Statistical Mechanics, 3rd ed. (2011), Ch. 14; Kardar, Statistical Physics of Particles (2007), Ch. 6
master: Pathria & Beale, Statistical Mechanics, 3rd ed. (2011), Ch. 14; Kardar, Statistical Physics of Particles (2007), Ch. 9; Chaikin & Lubensky, Principles of Condensed Matter Physics (1995), Ch. 7

References

Reif — Fundamentals of Statistical and Thermal Physics (McGraw-Hill, 1965) · Ch. 15 Fluctuations. Brownian motion and the Langevin equation, Einstein relation, correlation functions, Nyquist theorem for Johnson noise. Fluctuation-dissipation connection.
Pathria & Beale — Statistical Mechanics, 3rd ed. (Academic Press, 2011) · Ch. 14 Fluctuations. Fluctuation-dissipation theorem, Kubo formula, linear response theory, Onsager relations, correlation functions and spectral densities.
Kardar — Statistical Physics of Particles (Cambridge, 2007) · Ch. 6 Kinetic theory (fluctuation-dissipation), Ch. 9 Linear response and the Kubo formula. Langevin equation, Fokker-Planck equation, fluctuation-dissipation theorem and its derivation from linear response.
Einstein — Investigations on the Theory of the Brownian Movement (Ann. Phys. 17, 549, 1905) · Einstein 1905 paper on Brownian motion. Derived the relation between mean-square displacement and diffusion constant: ⟨x²⟩ = 2Dt, and connected D to temperature through D = kT/(6πηa).
Nyquist — Thermal Agitation of Electric Charge in Conductors (Phys. Rev. 32, 110, 1928) · Nyquist 1928 paper deriving the Johnson noise formula S_V = 4kTR from thermodynamic arguments and the second law.
Callen and Welton — Irreversibility and Generalized Noise (Phys. Rev. 83, 34, 1951) · Callen-Welton 1951 paper establishing the quantum-mechanical fluctuation-dissipation theorem, connecting dissipation to quantum fluctuations at finite temperature.
Kubo — Statistical-Mechanical Theory of Irreversible Processes I (J. Phys. Soc. Jpn. 12, 570, 1957) · Kubo 1957 paper introducing the general linear response formalism and the Kubo formula for transport coefficients as time-correlation integrals.

Estimated time

beginner: 15 min
intermediate: 40 min
master: 70 min