Stochastic simulations of fermionic dynamics with phase-space representations

A Gaussian operator basis provides a means to formulate phase-space simulations of the real- and imaginary-time evolution of quantum systems. Such simulations are guaranteed to be exact while the underlying distribution remains well-bounded, which defines a useful simulation time. We analyse the application of the Gaussian phase-space representation to the dynamics of the dissociation of an ultra-cold molecular gas. We show how the choice of mapping to stochastic differential equations can be used to tailor the stochastic behaviour, and thus the useful simulation time. In the phase-space approach, it is only averages of stochastic trajectories that have a direct physical meaning. Whether particular constants of the motion are satisfied by individual trajectories depends on the choice of mapping, as we show in examples.

💡 Research Summary

The paper presents a comprehensive framework for exact stochastic simulations of fermionic quantum dynamics using a Gaussian operator basis within a phase‑space representation. Traditional quantum Monte‑Carlo approaches often suffer from the sign problem and limited simulation times when applied to interacting fermion systems. By expanding the system’s density operator in terms of Gaussian operators—essentially non‑normalised coherent states—one can map the quantum evolution onto a set of stochastic differential equations (SDEs) for complex phase‑space variables. The resulting complex probability distribution remains bounded for a finite “useful simulation time,” guaranteeing that ensemble averages of the stochastic trajectories reproduce exact quantum expectation values throughout that interval.

A central contribution of the work is the systematic analysis of how the choice of mapping from the operator equations to the SDEs influences both the statistical properties of individual trajectories and the overall simulation time. The mapping determines how drift and diffusion terms are allocated among the phase‑space variables. By concentrating physically relevant information (e.g., conservation laws) in the drift while minimizing diffusion, one can extend the bounded‑distribution regime, thereby lengthening the useful simulation window. Conversely, increasing diffusion can improve the sampling of highly non‑linear interactions at the cost of larger statistical noise and a shorter simulation horizon. The authors demonstrate that different mappings can be tailored to prioritize specific observables—such as total particle number, energy, or mode occupations—without compromising the exactness of ensemble averages.

The methodology is applied to a concrete test case: the dissociation of an ultracold molecular gas into fermionic atom pairs. Initially the system consists of bound diatomic molecules; a sudden quench or external field drives the conversion into free fermions, a process characterized by rapid growth of correlations and non‑equilibrium population dynamics. Using the Gaussian phase‑space SDEs, the authors compute time‑dependent quantities including the average atom number, two‑point correlation functions, and total energy. Their results are benchmarked against conventional trajectory‑based methods, showing that for the same number of stochastic samples the phase‑space approach yields substantially lower statistical errors and maintains accuracy over a longer physical time before the distribution becomes unbounded.

Importantly, the paper emphasizes that while ensemble averages have direct physical meaning, individual stochastic trajectories need not satisfy conserved quantities exactly. Whether a trajectory respects a particular constant of motion depends on the chosen mapping. The authors illustrate this with examples where certain mappings enforce exact particle‑number conservation on each trajectory, whereas others only guarantee conservation on average. This flexibility allows researchers to design mappings that best suit the observables of interest or the computational resources available.

In conclusion, the Gaussian phase‑space representation offers a powerful, exact tool for simulating fermionic dynamics beyond the reach of traditional methods. By judiciously selecting the drift‑diffusion mapping, one can optimise the trade‑off between simulation length, statistical noise, and the fidelity of specific conserved quantities. The authors suggest that this approach can be extended to more complex fermionic systems—such as high‑temperature superconductors, quantum dot arrays, or nuclear matter—where strong correlations and non‑equilibrium effects are prominent. Future work will likely focus on multi‑mode extensions, stronger interaction regimes, and direct comparison with experimental data to further validate and refine the technique.