Variational Multi-Gaussian Quantum Trajectories

Technical Summary: Variational Multi-Gaussian Quantum Trajectories

Problem Statement
The theoretical investigation of open quantum many-body systems, particularly interacting bosonic systems, faces a formidable challenge due to the unbounded local Hilbert space. Exact simulations become rapidly intractable as occupation numbers and system sizes increase. While semiclassical approaches like the truncated Wigner approximation (TWA) and Gaussian approximations offer efficiency, they fail when interactions and non-Gaussian correlations become significant. Furthermore, while quantum trajectories provide a powerful framework for describing stochastic evolution conditioned on continuous monitoring, their exact simulation suffers from the same many-body complexity. There is a need for a method that can efficiently simulate stochastic quantum trajectories while capturing non-Gaussian physics beyond semiclassical limits.

Methodology: Stochastic Time-Dependent Variational Principle (sTDVP)
The authors propose a Stochastic Variational Multi-Gaussian (sVMG) framework. The core of this method is the generalization of the time-dependent variational principle (TDVP) to accommodate stochastic quantum state diffusion processes.

1. Stochastic Dynamics: The system is governed by a Lindblad master equation, which is "unraveled" into quantum trajectories using the heterodyne unraveling (Quantum State Diffusion). This leads to an Itô stochastic Schrödinger equation (SSE). To apply variational calculus, the authors convert this into an equivalent Stratonovich SSE, which preserves the ordinary chain rule required for projecting dynamics onto a variational manifold.
2. Variational Ansatz: The wavefunction $|\psi\rangle$ is approximated by a coherent superposition of $N_c$ Gaussian wavepackets:
   $$\langle x | \psi[\theta] \rangle = \sum_{k=1}^{N_c} G(x; \theta_k)$$
   Each Gaussian component is parametrized by a set of real parameters $\theta$ (including mean position/momentum, covariance matrices, and weights). For large systems, the covariance matrices can be compressed using a "low-rank" decomposition to reduce the number of parameters.
3. Equations of Motion: By projecting the exact stochastic increment of the wavefunction onto the tangent space of the variational manifold, the authors derive the sTDVP equation. This results in a set of stochastic differential equations for the parameters $\theta$:
   $$d\theta = \text{Re}(S^{-1}) \text{Re} \left( V dt + \sum_m Y^{(m)} \circ dZ_m^* \right)$$
   where $S$ is the quantum geometric tensor, and $V, Y^{(m)}$ are drift and noise force vectors. Crucially, because the ansatz consists of Gaussians, the integrals required to assemble $S$, $V$, and $Y^{(m)}$ (which involve polynomial-weighted Gaussian functions) can be computed fully analytically using Wick's theorem.

Key Contributions

• Generalization of TDVP: The work extends the time-dependent variational principle to stochastic dynamics driven by continuous measurement (quantum trajectories), rather than just unitary or Lindblad evolution.
• Analytical Assembly: The multi-Gaussian ansatz allows for the analytical derivation of the variational equations of motion, enabling efficient numerical integration without the need for Monte Carlo sampling of the wavefunction itself.
• Controlled Non-Gaussianity: The method systematically controls the non-Gaussianity of the wavefunction by varying the number of Gaussian components ($N_c$). As $N_c$ increases, the approximation converges toward the exact dynamics.

Results and Benchmarks
The method is benchmarked and applied to the driven-dissipative Bose-Hubbard model:

1. Bose-Hubbard Dimer ($N=2$):

   • The sVMG approach with a small number of components ($N_c=4$) faithfully reproduces exact individual trajectories and entanglement dynamics in regimes where semiclassical approximations (TWA, Gaussian Trajectory Approximation) fail.
   • The relative error in observables decreases systematically as $N_c$ increases.
   • The method successfully captures the symmetry-breaking phase transition and entanglement transitions in the thermodynamic limit (large $N$ scaling), matching mean-field predictions where non-Gaussianity diminishes.

2. Extended Bose-Hubbard Lattices (1D and 2D):

   • The framework is applied to 1D chains and 2D lattices to investigate symmetry-breaking transitions (checkerboard patterns).
   • Dimensionality Dependence: In 2D lattices, the sVMG results show the emergence of a clear symmetry-breaking phase with a distinct separation between checkerboard sublattices. In contrast, 1D chains exhibit local checkerboard-like regions separated by domain walls, resulting in a washed-out average order parameter. This suggests the symmetry-broken phase survives in 2D but is suppressed in 1D due to enhanced fluctuations.
   • The method captures these distinct behaviors using a low-rank ansatz with $N_c=4$ and rank $K=2$, demonstrating scalability beyond the reach of brute-force exact diagonalization.

Significance
The paper claims that the sVMG framework provides a versatile and efficient tool for simulating stochastic quantum dynamics in interacting bosonic systems. Its primary significance lies in its ability to bridge the gap between computationally cheap semiclassical methods and exact but intractable simulations. By capturing non-Gaussian correlations along individual trajectories, the method enables the study of phenomena such as entanglement dynamics and symmetry-breaking phase transitions in regimes where standard Gaussian or truncated Wigner approximations are insufficient. The ability to analytically assemble the equations of motion makes it a practical approach for exploring nonequilibrium phases in open quantum many-body systems.