Filtered Quantum Phase Estimation

Accurate state preparation is a critical bottleneck in many quantum algorithms, particularly those for ground state energy estimation. Even in fault-tolerant quantum computing, preparing a quantum state with sufficient overlap to the desired eigenstate remains a major challenge. To address this, we develop a unified framework for filtered-state preparation that enhances the overlap of a given input state through spectral filtering. This framework encompasses the polynomial and trigonometric realizations of filters, allowing a transparent analysis of the trade-offs between overlap amplification and preparation cost. As examples, we introduce signal-processing-inspired filters, such as Gaussian filters and Krylov subspace-based filters, that adaptively suppress excited-state contributions using low-rank projections. Within this framework, we further develop a filtered variant of QPE (FQPE) that mitigates the unfavorable dependence on the initial overlap present in standard QPE. Numerical experiments on Fermi-Hubbard models show that FQPE reduces the total runtime by more than two orders of magnitude in the high-precision regime, with overlap amplification exceeding a factor of one hundred.