Skip to content
Tutorials Center
My learning

Topics

Lesson 35 of 39 · Non-Equilibrium and Monte Carlo Methods

The Langevin Equation and the Fluctuation-Dissipation Theorem

Model a particle with friction plus random thermal kicks, derive equipartition and D = kT/gamma from the Langevin equation, and apply the fluctuation-dissipation idea to optical traps and Johnson noise.

Friction slows a moving particle down; thermal kicks keep it moving. Both come from the same source: collisions with the surrounding molecules. The Langevin equation (1908) puts them in one equation, and the fluctuation-dissipation theorem says the strength of the random force and the strength of the friction are not independent. Knowing one fixes the other. This lesson derives the essentials and simulates them.

The Langevin equation

For a particle of mass m and velocity v in a fluid with friction coefficient γ:

Text
m dv/dt = -γ v + ξ(t)

The first term is viscous drag. The second, ξ(t), is a rapidly fluctuating random force from molecular collisions, modeled as Gaussian white noise: zero mean and no memory,

Text
⟨ξ(t)⟩ = 0,         ⟨ξ(t) ξ(t')⟩ = 2 γ k T δ(t - t')

The coefficient 2γkT is the content of the fluctuation-dissipation theorem for this problem. It is the unique value for which the long-time state is the Boltzmann distribution at temperature T. You can see that it must tie γ to T: friction removes energy at a rate that grows with γ, and the noise must add energy at a matching rate, otherwise the particle would cool to rest or heat without limit.

Consequences:

  • Equipartition. In steady state ⟨½ m v²⟩ = ½ kT.
  • Velocity memory. ⟨v(0) v(t)⟩ = (kT/m) e^(-γt/m): velocities forget themselves in the time m/γ.
  • Green-Kubo relation for diffusion. D = ∫₀^∞ ⟨v(0) v(t)⟩ dt = kT/γ, the Einstein relation of the previous lesson.

White noise is an idealization: real collisions have a short but nonzero correlation time. It is accurate when you look at times much longer than the interval between molecular collisions.

Simulating the velocity

The code uses dimensionless units m = kT = γ = 1. The Euler-Maruyama update is v ← v - γ v dt / m + √(2 γ kT dt) N(0,1) / m. It measures the steady-state ⟨v²⟩, the velocity autocorrelation, and the diffusion constant from the position MSD.

Python Underdamped Langevin dynamics: equipartition and D = kT/γ

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The mean-square velocity is within about a percent of 1, and the measured autocorrelation follows e^(-γt/m). The diffusion estimate falls somewhat below 1 for two reasons: over a finite window the MSD is 2D(t - m/γ), a bias of about 5 percent at t = 20, and the single-time MSD from 1000 particles carries a statistical error of a few percent.

The overdamped limit and the optical trap

When friction is strong (small particles in liquids), inertia can be ignored over times longer than m/γ, giving γ dx/dt = F(x) + ξ(t). For a harmonic trap F = -κ x, the position is an Ornstein-Uhlenbeck process with relaxation time τ = γ/κ. Equipartition on the potential energy gives

Text
⟨x²⟩ = kT / κ,         ⟨x(0) x(t)⟩ = (kT/κ) e^(-t/τ)

This is how optical tweezers are calibrated in practice: track a trapped bead, then either compute the variance (κ = kT/⟨x²⟩) or the decay time (κ = γ/τ). For a bead of radius 0.5 µm in water (γ = 6πηa ≈ 9.4 × 10⁻⁹ kg/s) and a trap of stiffness κ = 10 pN/µm = 10⁻⁵ N/m, the relaxation time is about 1 ms and the rms position is about 20 nm. The simulation below uses the exact Ornstein-Uhlenbeck update, then "calibrates" the trap from the data.

Python Calibrating a simulated optical trap

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Both estimates return the true stiffness to within a few percent. This is a simulated bead; real experiments must also remove instrument noise and account for the medium's viscosity.

Johnson-Nyquist noise

The same theorem governs electrical noise. A resistor R at temperature T has a fluctuating open-circuit voltage with mean-square value

Text
⟨V²⟩ = 4 k T R Δf

in a frequency band Δf. The resistance is the dissipation; the voltage noise is the fluctuation. Likewise, a capacitor C in thermal equilibrium has ⟨V²⟩ = kT/C (equipartition for the energy ½ C V²).

Python Thermal noise of a resistor and a capacitor

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The 1 MΩ resistor produces about 13 µV rms in 10 kHz: small, but it sets the limit on how weak a signal an amplifier with that input resistance can detect.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. What does the fluctuation-dissipation theorem say about the Langevin noise strength?

    ⟨ξξ⟩ = 2γkT δ(t - t'); only this value yields the Boltzmann distribution.
  2. 2. In the Langevin model, the steady-state average kinetic energy of the particle is

    That is equipartition; the drag and noise balance exactly at this value.
  3. 3. A bead is trapped with stiffness κ. Its position variance ⟨x²⟩ is

    Equipartition on ½ κ x² gives ½ κ⟨x²⟩ = ½ kT.
  4. 4. If you double the resistance R, the Johnson noise voltage (rms, fixed band) grows by a factor of

    ⟨V²⟩ ∝ R, so V_rms ∝ √R.

Summary

  • Langevin: m dv/dt = -γ v + ξ(t) with ⟨ξ(t)ξ(t')⟩ = 2γkT δ(t - t').
  • It reproduces equipartition, velocity decay e^(-γt/m) and D = kT/γ.
  • Overdamped harmonic traps give ⟨x²⟩ = kT/κ and relaxation time γ/κ.
  • Johnson noise ⟨V²⟩ = 4kTRΔf is the electrical version of the same fluctuation-dissipation balance.

Next: Practice Set 1: Ideal Gases and Quantum Statistics