Package to simulate coupled exciton and photon driven-dissipative Gross–Pitaevskii equations with a dark excitonic reservoir, using a first-order split-step spectral scheme and CuPy.
Clone the repository and install the package:
git clone https://github.com/Quantum-Optics-LKB/ddGPE.git
cd ddGPE
pip install .Running simulations requires a working CuPy installation and a compatible CUDA GPU.
The fields are evolved in the frame rotating at the pump angular frequency, with energies divided by hbar. With apply_reservoir=True, the deterministic model is
Here the bare decay rates gamma_X and gamma_C refer to constructor inputs, before conversion losses are added. The linear propagation uses the Hopfield approximation described below.
| Symbol | Code parameter or array | Meaning |
|---|---|---|
|
|
phi[0], phi[1]
|
Exciton and photon fields |
den_reservoir |
Reservoir density, not a field amplitude | |
g0 |
Exciton interaction coefficient; also used for the reservoir-induced exciton shift | |
rabi |
Off-diagonal exciton–photon coupling | |
|
|
Input gamma_exc, gamma_cav
|
Bare coherent density decay rates |
gamma_res |
Coherent-to-reservoir conversion rate | |
gamma_res_decay |
Reservoir decay rate, set to the original input gamma_exc
|
|
potential |
Cavity potential energy divided by hbar | |
v_gamma |
Additional photon density loss at the boundaries |
The reservoir shifts only the exciton energy directly. Its effect on photons follows through exciton–photon coupling. There is no reservoir-dependent coherent gain, reservoir diffusion, or exciton potential in this implementation.
Conversion adds gamma_res to both coherent decay rates. The reservoir decay remains equal to the original gamma_exc input:
self.gamma_exc = input gamma_exc + gamma_res
self.gamma_cav = input gamma_cav + gamma_res
reservoir decay = input gamma_exc
Thus conversion removes coherent density and supplies reservoir density at matching rates in the differential model.
With apply_reservoir=False, the extra coherent losses and reservoir update are disabled. The nonlinear kernel still uses the supplied reservoir density: a nonzero initial_state[2] acts as a frozen exciton shift. The default initial reservoir is zero.
The exciton dispersion is flat, while the photon dispersion is
The pump frequency is defined using the bare, lossless lower-polariton energy at zero momentum:
Positive detuning places the pump above this reference. For a finite pump wavevector, the effective detuning from the LP resonance also includes the dispersion shift at that wavevector.
The cavity drive is the sum of pump and probe amplitudes:
F = F_pump * F_pump_r * F_pump_t + F_probe * F_probe_r * F_probe_tThese profiles multiply field amplitudes; intensity is proportional to the squared modulus. A probe factor exp(-1j * omega_probe * t) denotes a frequency offset above the pump for positive omega_probe.
| Quantity | Units |
|---|---|
| Time | ps |
| Position | micrometres |
| Fields | inverse micrometres |
| Field and reservoir densities | inverse square micrometres |
omega_exc, omega_cav, rabi, detuning, decay rates |
inverse ps (angular-frequency convention) |
g0 |
square micrometres per ps |
potential |
inverse ps |
| Drive amplitude, for dimensionless profiles | inverse micrometres per ps |
| Wavevector | inverse micrometres (angular wavevector) |
Convert energies in meV to evolution coefficients by dividing by hbar in meV ps. A physical potential V must be supplied as V / hbar; an interaction energy shift is hbar * g0 * density.
The term -1j * gamma / 2 gives amplitude decay exp(-gamma*t/2) and density decay exp(-gamma*t). The corresponding isolated-mode energy linewidth (FWHM) is hbar * gamma.
The lossless resonant Rabi splitting is 2 * hbar * rabi. In the thesis convention with off-diagonal coupling hbar * Omega_R / 2, use rabi = Omega_R / 2.
For a predominantly LP field at low momentum, let X02 = |X_0|^2 be its exciton fraction and C02 = 1 - X02 its photon fraction. The code uses the phase convention
psi_LP = -X * psi_X + C * psi_C
psi_UP = C * psi_X + X * psi_C
Projection with approximately constant Hopfield coefficients gives
g_LP = g0 * X02**2
g_res_LP = g0 * X02
F_LP = sqrt(C02) * F
bare gamma_LP ≈ X02 * input gamma_exc + C02 * input gamma_cav
additional gamma_LP = gamma_res
The model fixes the reservoir coupling and decay through these relations; they are not independent constructor inputs.
At steady state in this approximation:
n_res / n_LP = gamma_res / input gamma_exc
beta = (g_res_LP * n_res) / (g_LP * n_LP)
= gamma_res / (X02 * input gamma_exc)
examples/physical_constants.py targets beta = 0.49 by calculating the zero-momentum exciton fraction and setting
delta_cx = omega_cav - omega_exc
X02 = 0.5 * (1 + delta_cx / (delta_cx**2 + 4 * rabi**2)**0.5)
gamma_res = 0.49 * X02 * gamma_excThis is a steady-state, low-momentum calibration, not a constraint on transient or strongly mixed LP/UP states.
These conventions correspond to the LP reservoir model in Eq. (1.91) of K. Guerrero-Feuillet, Rotational superradiance in a polariton quantum fluid (2026), with thesis gamma_in mapped to code gamma_res, and thesis reservoir decay mapped to the input gamma_exc. The measured reservoir blueshift ratio is discussed in thesis section 3.1.7.
- The linear step uses complex polariton eigenfrequencies with lossless Hopfield coefficients. This neglects the small change in composition due to unequal component losses; it becomes exact for equal losses.
- The solver retains both branches and the full photon dispersion. The scalar LP model follows in the low-momentum, predominantly LP regime.
- The reservoir evolves dynamically. Comparing with the thesis Bogoliubov treatment, which neglects reservoir-density perturbations, requires that additional approximation.
- The equations above describe deterministic evolution. Optional noise is added after
t_noise; its current increments scale withdtrather thansqrt(dt), so it is not a correctly time-step-scaled continuous white-noise discretization.