arXiv:2609.18246
New directions in dynamical expectation estimation
Panjin Kim, Kyung Chul Jeong, Yun-Tak Oh
Abstract
Computing dynamical expectation values typically relies on approximations optimized for the state or observable separately, without accounting for how their errors combine in the final expression. This work explores an alternative approach in which each approximation is guided by both the state and the observable, accounting for how they jointly determine the target expectation value. This idea is implemented through a sweep algorithm with coupled loss functions for forward state and backward observable updates. Exact error relations provide an analytical rationale for how the proposed losses can improve the accuracy. Numerical tests on 30-qubit random circuits show errors two to three orders of magnitude smaller than those of variational state compression at equal bond dimensions. These results motivate further exploration of joint state and observable approximation for dynamical expectation value estimation.
AI Summary
Standard tensor-network simulations truncate states or observables separately, although the target is often a single expectation value. This paper investigates whether conditioning each approximation on both objects can reduce the final estimation error. The proposed algorithm sweeps forward through a circuit to update approximate states, then backward to update approximate observables. Each compression loss combines a Hilbert–Schmidt distance penalty with a penalty on the change in the state–observable contraction. The implementation uses matrix-product representations and alternating local tensor updates. Exact error identities decompose the final error into contraction-weighted backward compression defects and a remainder involving products of state and observable errors. The contraction penalty directly targets the first term. If accumulated state errors and local backward defects are both $O(\epsilon)$, the remainder is bounded by $O(L\epsilon^2)$ for $L$ evolution steps; this is a conditional scaling argument, not a guaranteed improvement. Numerical benchmarks use 30-qubit random barren-plateau circuits with 5–100 layers, averaging over 500 instances and estimating $\langle Z_1Z_2\rangle$ against exact state-vector results. With equal forward and backward bond dimensions and loss-weight ratio $\gamma/\mu=1000$, reported errors are two to three orders of magnitude below variational state compression at the same bond dimension. Pauli propagation is also benchmarked. The metric is relative error capped at one. The loss functions are explicitly heuristic, and neither their optimal form nor the optimal realization of joint approximation is established. The implementation considers nearest-neighbor two-qubit gates, and the numerical evidence concerns the tested circuit ensemble and observable.
Error found: The forward ket-MPS implementation claims linear local stationarity equations, although both terms of its stated density-operator loss are generally quartic in the updated ket tensor, leaving the relationship between the implemented updates and the proposed optimization unclear.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 12 pages, 3 figures