arXiv:2610.20791
Sparse-QITE: Discovering Compact Generators for Quantum Imaginary-Time Evolution via Regularization Methods
Rafael Gómez-Lurbe, Luca Ion, Alberto Acevedo, Andreu Anglés-Castillo
Abstract
Quantum imaginary-time evolution (QITE) provides a systematic route to ground-state preparation, but its cost grows rapidly with the locality $D$ of the unitary approximation, for which the local Pauli basis scales as $4^D$. We introduce Sparse-QITE, a regularized formulation that promotes compact generators, reducing the number of quantum operations and yielding circuits with fewer gates and lower depth. We derive rigorous bounds on the sparsification error and benchmark the method in one-dimensional transverse-field Ising and antiferromagnetic Heisenberg models. Sparse-QITE exhibits a controllable sparsity--accuracy trade-off while retaining strong compression even in the high-accuracy regime. In scalability experiments with respect to $D$, the active support grows much more slowly than in standard QITE and, in several regimes, approximately saturates over the domain sizes studied. After circuit construction and transpilation, some of the largest-domain cases require only about $1\%$ of the standard-QITE gate count while maintaining high accuracy. We further introduce adaptive and active-discovery variants: the former selects the regularization strength dynamically during the evolution, while the latter reduces the number of Pauli expectation values that need to be evaluated and, consequently, the measurement requirements. These results establish sparsity-promoting regularization as a route toward more resource-efficient QITE. More broadly, these results suggest that sparsity-promoting regularization may also benefit variational quantum algorithms.
AI Summary
QITE approximates local imaginary-time updates with unitary generators expanded in up to $4^D$ Pauli strings. This paper asks whether sparse generators can reproduce QITE trajectories while reducing implementation costs. Sparse-QITE adds an $\ell_1$ penalty to the quadratic coefficient objective, with a small ridge term for stability, and solves the convex problem using FISTA. The implementation also refits coefficients on the selected support to reduce shrinkage bias. Benchmarks use eight-spin periodic transverse-field Ising chains in ordered and disordered regimes and an antiferromagnetic Heisenberg chain. At final fidelity at least $0.999$ with standard QITE, representative Ising runs retain $3.03\%$ and $16.0\%$ of its accumulated generator terms, respectively. Domain-size sweeps over $D=2,\ldots,7$ show slower support growth than the full pool and approximate saturation in some regimes. Suzuki decomposition and transpilation yield gate counts around $1\%$ of standard QITE in some large-domain cases. Adaptive regularization controls the local objective mismatch; a heuristic discovery-and-freezing variant reduces subsequent expectation-value requirements. For pool size $M$ and minimum real-Gram eigenvalue $\kappa$, the sparse-versus-ridge tangent error is bounded by $\frac{\lambda}{2}\sqrt{M/(\kappa+\rho)}$. A trajectory bound is also derived, but its exponential stability factor makes it quantitatively uninformative at the numerical ridge value $\rho=10^{-10}$. The basic method retains full-pool measurement costs. The experiments establish no asymptotic scaling law; support freezing has no guarantee against missing later operators, and adaptive binary search assumes mismatch monotonicity not guaranteed with ridge regularization. Larger-system behavior and realistic-noise performance remain open.
Error found: The ordered-phase λ=1 table row reports zero implemented terms but a ground-state fidelity of 1.36×10^-7, whereas the declared unchanged initial state has fidelity approximately 0.436 for the stated Hamiltonian. The implementation debiases elastic-net solutions, but the paper does not establish its adaptive-search justification for these debiased coefficients, although some local error bounds can be extended. The paper defines active support as a union of distinct Pauli strings but reports resource counts that include repeated occurrences across separate local generators.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.