arXiv:2610.02082
Strong Simulation of 1D Quantum Circuits via Reduced Transition Matrices
Matilde Grassi, Stefano Carignano, Luca Tagliacozzo, Jacopo De Nardis
Abstract
Tensor networks are powerful tools for simulating quantum many-body systems, but the growth of spatial entanglement severely limits the direct evolution of pure states under chaotic unitary circuits. Here we consider a more targeted task: the approximate strong simulation of a specified output probability of a 1D chaotic brick-wall circuit at finite relative precision. Given input and output bit strings $\by$ and $\bx$, we evaluate $p(\bx|\by)=|\bra{\bx}U(T)\ket{\by}|^2$ using the Sweeping RTM algorithm, a transverse tensor-network contraction based on reduced transition matrices (RTMs). The algorithm compresses the left and right temporal boundary states jointly, seeking an output probability that converges across spatial cuts and as the bond dimension is increased. For chaotic one-dimensional brick-wall circuits at a fixed relative target precision, we find numerical evidence that the bond dimension required to obtain stable estimates grows subexponentially over the accessible time window. Our findings open a direct route to classical probability queries for chaotic quantum circuits, with potential applications to benchmarking and learning tasks.
AI Summary
The paper studies approximate strong simulation of a specified output probability $p(\mathbf{x}|\mathbf{y})=|\langle\mathbf{x}|U(T)|\mathbf{y}\rangle|^2$ of a chaotic one-dimensional brick-wall circuit at fixed relative precision, rather than sampling its output distribution or representing the full evolved state. The proposed Sweeping RTM algorithm contracts the circuit tensor network transversely. It jointly compresses left and right temporal matrix-product states using singular subspaces of their reduced transition matrices, targeting their overlap rather than either boundary separately. Alternating spatial sweeps and increasing bond dimension are checked for agreement across cuts, successive sweeps, and bond dimensions. Numerical tests cover a deterministic kicked-Ising circuit, exchange-symmetric random gates, and Haar-random gates. For $N=60$ and amplitude tolerances $\epsilon_{\mathcal A}=0.8,0.5,0.1$, the required bond dimensions show growth compatible with subexponential behavior over the accessible time window; Haar circuits are substantially harder. Comparisons with exact evolution at $N=20$ find mean relative amplitude errors below the chosen tolerances. Larger-system checks use cross-tolerance agreement and convergence of mean log probabilities toward Haar predictions. Retained RTM spectra exhibit approximately exponential tails, although generic finite-chain entropy data do not establish logarithmic growth. The evidence uses one circuit realization per family and up to 40 input-output pairs. No asymptotic complexity bound or rigorous relative-error guarantee is established, and sweep convergence is not proved. The paper discusses conditional applications to cross-entropy benchmarking, phase-sensitive fidelity certification, and likelihood-based learning; extending the method to two dimensions remains nontrivial.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- comments welcome