Publication Deep Dive · 2025

Quantum Dynamics Simulations of Exciton Polariton Transport

B. X. K. Chng, M. E. Mondal, W. Ying, and P. Huo · Nano Letters 25, 1617–1622

Paper ↗PDF ↗Code ↗All publications
A polariton wavepacket spreading through a multimode cavity and leaving a localized dark-state wake

From the paper

Simulating transport at the spatial scale of experiments

Experiments image polariton motion over micrometers, yet measured wavefront velocities are often below the derivative of the ideal lower-polariton dispersion. The calculation must include thousands of molecular sites, hundreds of dispersive photon modes, local phonon environments, and photon loss. The paper develops a method that retains this spatial structure without storing the dense exciton–photon coupling matrix.

From the paper

Generalized HTC Hamiltonian

\[\hat H_{LM}=\sum_{k,n}g_k\left(\hat a_k^\dagger\hat\sigma_n e^{-ikx_n}+\hat a_k\hat\sigma_n^\dagger e^{ikx_n}\right).\]

The finite-$k$ phase factors couple a real-space molecular array to a photon momentum grid. Each exciton has a classical Debye bath propagated with mean-field Ehrenfest forces, and every cavity mode undergoes phenomenological Lindblad loss. Phase-matched bright excitons form momentum-resolved upper and lower polaritons; molecular Fourier components without a matching photon remain dark.

From the paper

Why the FFT changes the feasible system size

The exciton–photon block is a discrete Fourier transform. Applying it with FFTs changes the dominant action from a dense quadratic operation to approximately $O(N\log N)$. The production workflow then:

  1. Build the dispersive photon grid and uniformly spaced molecular array.
  2. Prepare a localized polariton wavepacket from an energy-selected band.
  3. Sample independent molecular baths from a thermal Wigner distribution.
  4. Propagate quantum amplitudes with RK4 and the FFT Hamiltonian action.
  5. Advance classical bath coordinates with velocity Verlet.
  6. Apply stochastic photon-loss steps and average many trajectories.
  1. Real spaceLocalized molecular amplitudes and bath coordinates.
  2. FFT actionTransform only the phase-matched components.
  3. Momentum spaceDispersive photons and polariton propagation.

From the paper

Transport is more than a group velocity

\[\sigma^2(t)=\langle\psi(t)|(\hat x-\langle x\rangle)^2|\psi(t)\rangle,\qquad \sigma^2(t)=\sigma^2(t_0)+Dt^\gamma.\]

The wavefront position gives a velocity, while the transient mean-square displacement diagnoses ballistic ($\gamma=2$), diffusive ($\gamma=1$), and subdiffusive behavior. Decomposing the density into UP, LP, dark, and interference contributions explains why the upper-polariton MSD can rise and then fall.

Simplified educational example

Bath-free polariton dispersion and a phenomenological transport guide

For an exciton energy $E_{ex}$, dispersive photon energy $E_c(k)$, and collective coupling $G$, diagonalizing the two-state block at each momentum gives:

\[E_\pm(k)=\frac{E_{ex}+E_c(k)}2\pm\sqrt{\left(\frac{E_{ex}-E_c(k)}2\right)^2+G^2}.\]

The numerical derivative $\hbar^{-1}\partial E_-(k)/\partial k$ is the ideal lower-polariton group-velocity guide. The paper's simulated wavefront can be slower because bath coupling and photon loss reshape the packet.

Mechanism-level teaching curves. They are not trajectories from the L-MFE calculation.
Python polariton dispersion
import numpy as np

hbar_c = 0.1973269804   # eV micrometer
hbar = 0.0006582119569  # eV ps
Ec0, Eex, G = 1.90, 1.996, 0.120
k = np.linspace(-12.0, 12.0, 4001)
Ec = np.sqrt(Ec0**2 + (hbar_c*k)**2)
split = np.sqrt(((Eex-Ec)/2)**2 + G**2)
Elp = (Eex+Ec)/2 - split
vlp = np.gradient(Elp, k) / hbar
print(vlp[np.argmin(abs(k-5.0))], "micrometer/ps")

From the paper

Microscopic picture of the reported transport

  • Exciton–phonon coupling and photon loss both reduce the observed wavefront velocity below the ideal dispersion slope.
  • A rapidly moving UP packet deposits dark population at positions it has already visited; that dark wake is immobile in the dispersionless bare-exciton model.
  • UP matter–photon interference is constructive and disappears with decoherence, contributing to a pronounced MSD peak and decline.
  • The LP expands more gradually and approaches a plateau rather than showing the same collapse.
  • The fitted MSD exponent crosses from ballistic toward diffusive, with more photonic packets remaining ballistic for longer.

Scope and limitations

What is simplified

  • The molecular baths are classical and propagated by mean-field Ehrenfest dynamics.
  • No static disorder, direct excitonic coupling, or Peierls coupling is included.
  • The bare/dark exciton band is therefore dispersionless.
  • Molecules are identical, periodic, and restricted to the single-excitation manifold.
  • The loss rate is common to all photon modes, and extracted velocities depend on the chosen observable and fitting window.