arXiv:2610.29372
Quantum Simulation of Non-Markovian Open Quantum Systems via Low-Rank Kernel Compression and Coupling Dilution
Hengzhun Chen, Zhenning Cai
Abstract
We present an error-controlled framework for the quantum simulation of non-Markovian open-system dynamics in Gaussian environments. For a system coupled linearly to a Gaussian environment with zero mean, the bath's influence on the reduced dynamics is completely determined by its two-time correlation kernel. This allows us to replace the microscopic environment with an equivalent bath that has the same kernel. On a finite simulation interval, a rank-$N$ approximation of the kernel yields an effective bath of $N$ bosonic modes with time-dependent couplings to the system. To encode these modes in qubits, we first truncate each mode to its lowest $d$ occupation levels. When stronger couplings lead to large occupations, we introduce coupling dilution, which replaces each mode with several more weakly coupled copies while preserving the bath correlation kernel. This distributes excitations across more modes and trades additional modes for smaller local dimensions. We derive explicit reduced-state error bounds and establish polynomial circuit width and depth in simulation time and inverse accuracy under stated spectral-decay and coupling-regularity assumptions, provided elementary unitaries admit polynomial-cost implementations.
AI Summary
The paper addresses how to represent environmental memory with finite quantum resources when simulating a system linearly coupled to an initially uncorrelated, zero-mean Gaussian bosonic bath. It constructs an auxiliary vacuum bath directly from the two-time correlation kernel $G(t,s)$, including nonstationary kernels, rather than discretizing the microscopic environment. On $[0,T]$, retaining $N$ kernel eigenpairs defines time-dependent couplings $g_n(t)=\sqrt{\lambda_n}u_n(t)$. The reduced-state trace-norm error is bounded by $\exp[2\|S\|^2T(\sum_{n>N}\lambda_n^2)^{1/2}]-1$. Each auxiliary oscillator is then truncated to $d$ levels. Alternatively, “coupling dilution” replaces each mode by $q$ copies coupled with amplitude $g_n/\sqrt q$, preserving the Gaussian kernel before occupation truncation. With $\beta_n=\|S\|\int_0^T|g_n(t)|dt$, the diluted occupation error is at most $2^dq^{1-d}\sum_n\beta_n^{2d}$; two-level copies therefore give $4\sum_n\beta_n^4/q$. The bounds use Gaussian correlation equivalence, a correlation-perturbation theorem, and Duhamel expansions that track occupation-boundary crossings and cancellations under the bath trace. A midpoint–Strang circuit has error $T(\Delta t)^2R$, with vacuum-dependent occupation estimates avoiding artificial dilution dependence in two contributions to $R$. Polynomial circuit width and depth in $T$ and inverse accuracy follow under specified eigenvalue-decay and coupling-regularity assumptions, conditional on polynomial-cost elementary unitaries. The analysis fixes the system operators and treats one coupling operator. Numerical kernel reconstruction and gate synthesis require separate error budgets. Dilution trades more registers for smaller local dimensions without guaranteeing lower depth; representative numerical benchmarks and multiple-coupling extensions remain future work.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.