Quantum algorithms gain from filtered-state preparation

Researchers from Sungkyunkwan University in South Korea and Xanadu in Canada detailed a new method for improving quantum algorithms in a paper published September 25, 2026, in Quantum Science and Technology.

The team developed a technique designed to address a critical limitation in many quantum algorithms: accurate state preparation. This work demonstrates a framework for enhancing the overlap of input states through spectral filtering, potentially reducing runtime by more than two orders of magnitude for certain calculations, with overlap amplification exceeding a factor of one hundred.

Filtered Quantum Phase Estimation for Eigenvalue Problems The success probability of quantum phase estimation hinges on initial state overlap; a new framework detailed in Quantum Science and Technology on September 25, 2026, directly addresses this limitation through filtered-state preparation. Researchers from Sungkyunkwan University and Xanadu developed a method to amplify this overlap, potentially reducing the computational cost of determining essential properties of many-body Hamiltonians, such as ground-state energy and excited spectra. The work introduces a unified framework for quantum algorithms centered on enhancing the initial connection with target eigenstates. This framework explicitly defines the trade-off between overlap amplification, the probability of successfully preparing a state, and the resources required to implement the filter itself. Analysis of Gaussian filters and a modified Krylov-subspace-based filter revealed improvements in the success-probability/overlap balance important for preparing states.

The team’s approach tackles a central challenge in quantum computing: accurately estimating eigenvalues, a task where standard quantum phase estimation requires a significant initial overlap between the prepared input state and the target eigenstate. Specifically, the success probability of QPE scales with the squared overlap, meaning even small improvements in initial overlap can yield substantial gains. The researchers formulated a variational problem leading to a generalized eigenvalue equation, with the lowest generalized eigenvector defining the standard Krylov filter. However, they found that while the Krylov filter accurately estimates energy, it can be suboptimal for filtered-state preparation; suppressing excited-state components too strongly can reduce the overall filter amplitude and increase postselection overhead. “Thus, the KSD eigenvector can be suboptimal for FQPE even when it gives an accurate variational energy,” the paper states, highlighting the need for a more subtle approach. The normalization of implemented filters directly impacts postselection success probability, calculated as pf–1, which must be considered alongside the overall computational cost. In testing, filtering consistently increased the effective overlap with the target eigenstate and improved the success probability of QPE; Gaussian filters demonstrated robust performance with a limited number of basis terms, while Krylov filters exhibited strong localization properties with shallow-depth implementations. This suggests practical relevance for the proposed framework, offering a pathway to more efficient eigenvalue estimation tasks and potentially accelerating progress in fields reliant on accurate quantum simulations. The rapid acceptance of the paper, following submission on February 5, 2026, and revisions completed by June 12, 2026, may indicate a fast-moving area of research within quantum computing.

Initial State Overlap as a Bottleneck in Quantum Phase Estimation Preparing an initial quantum state with sufficient similarity to the desired solution remains a primary obstacle in many quantum algorithms, even with anticipated advances in fault-tolerant computing. This approach moves beyond simply assuming a good initial state or applying spectral transformations without accounting for implementation demands. This filter achieved the lowest quantum phase estimation cost when compared to other approaches tested, suggesting a pathway to more efficient eigenvalue calculations.

The team’s work positions filtered-state preparation as a broadly applicable method for enhancing quantum eigenvalue estimation algorithms, addressing a fundamental limitation in accurately determining properties like ground-state energy and excited spectra. However, generating states with significant overlap is often the primary challenge in quantum simulation, particularly for complex systems like large molecules or strongly correlated electron models. When this initial overlap is minimal, represented as |γ0|^2≪1, quantum phase estimation demands numerous repetitions or postselections, escalating the computational burden even on future fault-tolerant devices. “As a result, the modified Krylov filter achieves the lowest FQPE cost among the three approaches,” the paper states, highlighting the efficiency gains achieved through spectral filtering.

Exponential Overlap Decay and the Orthogonality Catastrophe Even physically realistic input states suffer diminished overlap with true ground states as system size increases, a consequence of sensitivity to even minor Hamiltonian perturbations mirroring the effects of electron correlation. This inherent instability, termed the orthogonality catastrophe, presents a significant hurdle for quantum algorithms reliant on accurate initial state preparation, as even slight deviations drastically reduce fidelity. The research details how manipulating the initial overlap, represented as |γ₀|², becomes critical when tackling complex quantum simulations.

The team explored a method to intentionally alter this overlap through the application of a, a bounded mathematical operation on the Hamiltonian, creating a filtered state |φf⟩. While this manipulation introduces a postselection overhead, quantified as pf⁻¹, and demands a certain filtering depth D_(sp, f), the potential gains in algorithm efficiency are substantial. Limitations of Existing State Preparation Methods Classical methods for preparing quantum states face inherent limitations, with techniques like truncated configuration interaction struggling to accurately model many-body correlations and incurring substantial gate overhead when translated into quantum circuits. Adiabatic continuation, while conceptually straightforward, is hampered by the spectral gap bottleneck, where the time needed for reliable evolution diminishes rapidly as the minimum gap closes, increasing susceptibility to decoherence. Tensor network-based approaches, reliant on classical optimization, become computationally prohibitive in complex, high-dimensional systems. The research details a departure from simply extracting eigenvalues, instead focusing on directly preparing high-fidelity ground states through a tailored application of the Krylov subspace determinant (KSD) algorithm. This integration into a filtering-based framework allows for a more flexible adaptation to a Hamiltonian’s spectral structure, selectively suppressing contributions from excited states and accelerating convergence. Unlike fixed-shape filters that adjust only peak location and width, this method reshapes the spectral weight of the initial state in a controlled manner, complementing existing polynomial and Fourier-based Hamiltonian transformations. This framework treats filtering as a distinct algorithmic layer designed to increase the effective overlap with the desired eigenstate, thereby reducing the overall cost of subsequent algorithms. The potential benefit of this approach, however, is not guaranteed to scale indefinitely; the research indicates that the advantage of filtered quantum phase estimation can diminish if the physical gap between ground and excited states closes too quickly.

The team examined Krylov-based filter constructions as an alternative to analytically defined filters, tailoring the modified Krylov ansatz specifically for state preparation rather than solely eigenvalue estimation. “By adjusting the construction to trade off spectral leakage against success probability, Krylov filters can achieve strong selectivity with shallow circuits and a small number of basis states,” the paper states. The principles explored are not limited to ground-state preparation; the same techniques can be extended to excited states through appropriate adjustments to the filter function. Future research will explore these extensions, as well as connections to hybrid and sample-based state preparation methods, potentially broadening the applicability of this approach. The data supporting these findings are openly available, allowing for independent verification and further development of the techniques. The authors acknowledge Youngjun Park for helpful discussions during the research process. Cost-Aware Framework for Filtered-State Preparation The work introduces a cost-aware approach to filtered-state preparation, designed to improve the efficiency of ground-state energy estimation even within the constraints of fault-tolerant quantum computing. This framework allows for a tunable trade-off between convergence sharpness and success probability, a key factor in scaling quantum algorithms to larger, more complex systems. The research team addressed the typical exponential decay in success probability seen when sharpening Krylov filters by developing a modified KSD protocol. This protocol avoids the need for precise prior estimates of ground-state energy or energy gaps, streamlining the process of setting filter parameters. The modification maintains the computational structure of standard Krylov subspace decomposition, meaning existing computational tools can be readily adapted. After constructing matrices H and S, filter coefficients are still derived from a generalized eigenvalue problem, simplifying implementation. “The penalty becomes stronger when filtered-state preparation is expensive compared with the subsequent QPE circuit, and weaker when high-precision QPE dominates the cost,” the paper states, detailing the framework’s adaptability. The framework’s versatility extends beyond simply improving performance with existing methods. It provides a unified approach to both polynomial and trigonometric realizations within quantum phase estimation. The paper’s organization reflects this comprehensive approach, dedicating sections to defining filtered states, formulating the FQPE cost theorem, analyzing Gaussian FQPE and introducing the modified Krylov filters. This systematic exploration positions the work as a foundational contribution to the field, offering a cost-aware strategy for optimizing quantum algorithms and accelerating progress toward practical quantum computation. Polynomial and Trigonometric Filter Realizations The proposed framework accommodates both polynomial and trigonometric realizations of filters, explicitly quantifying the balance between amplifying overlap, achieving preparation success and the cost of implementing the filter itself. This versatility extends to handling diverse filter constructions, with the technical details of Hamiltonian function construction using QSVT, QETU, and GQSP detailed in supplementary materials. The core principle involves a filter function reshaping spectral amplitudes, suppressing unwanted components while preserving the target eigenstate. Implementing a Gaussian filter can use either trigonometric expansions or polynomial approximations, both exhibiting comparable asymptotic cost scaling; the number of basis functions required scales similarly for both approaches. Specifically, the filtering depth scales as Dsp,g = O(Δ E_0^(-1)) with parameter choices outlined in equation (15) within the published paper. This cost scaling is significant because it suggests a pathway to practical implementation even with limited quantum resources, a persistent challenge in the field. The framework’s ability to handle both polynomial and trigonometric realizations provides flexibility in adapting to specific hardware constraints and algorithmic requirements. Krylov filtering distinguishes itself by constructing the filtered state within a variational subspace generated by the input state and Hamiltonian basis functions. This approach dynamically adjusts filter coefficients based on the input state’s spectral distribution. This allows for direct incorporation of the cost-overlap trade-off into the state preparation process, a feature not typically found in static filter designs. The basis functions employed can be either polynomial, such as xk or Tk(x), or trigonometric, like e^(i kπ x), offering further adaptability. Krylov-Subspace Filter Improves Success Probability The modified Krylov filter addresses a key limitation of standard Krylov approaches by introducing a tunable parameter, Λ, that balances overlap amplification with post-selection success probability, incurring minimal classical computational overhead. This adjustment allows for a more robust and reliable filtered-state preparation process, important for complex quantum computations. Equations presented in the paper combine the filter’s success probability and the resulting target-state overlap, revealing a relationship where these two factors cannot be independently maximized. Specifically, the research shows that a filter enhances the input state when it preserves the desired target component while suppressing unwanted spectral weight, a condition defined by the inequality presented in the paper. Standard QPE, when initialized with a given input state, estimates an eigenvalue with a certain failure probability using a defined total depth. Conversely, FQPE, using the filtering technique, achieves the same estimation with an equivalent failure probability but uses a potentially reduced total depth, suggesting a pathway to more efficient algorithms. Numerical results further illustrate this advantage, contrasting the standard Krylov filter with the modified version, and demonstrating how the latter maintains success probability while increasing overlap. Curves within the paper depict the relationship between Krylov dimension, overlap, success probability, and relative query cost, revealing the benefits of the modified approach for achieving target accuracies of 10–4Δ E0. Runtime Reduction of Filtered QPE on Fermi-Hubbard Models Numerical simulations of Fermi-Hubbard models reveal that filtered quantum phase estimation, a technique for refining quantum algorithm efficiency, decreases total runtime by more than two orders of magnitude when high precision is required, with improvements to overlap exceeding a factor of one hundred. This performance gain stems from a refined approach to state preparation, focusing computational resources on the most relevant quantum states. Researchers explicitly compared the performance of Gaussian and Krylov filters, assessing their effectiveness in preparing the necessary quantum states for accurate eigenvalue estimation. The efficiency of this filtered approach is measured by comparing the expected total circuit depth required for standard quantum phase estimation versus the filtered method, considering both the preparation of the initial state and the subsequent eigenvalue estimation.

The team normalized resulting Hamiltonians, mathematical descriptions of the system’s energy, to ensure their spectra lie within a defined range of -1 to 1, a standardization that simplifies the computational process. This normalization involves partitioning Pauli operators, fundamental building blocks of quantum information, into a linear combination of unitary operators, a technique that allows for more efficient calculations. The resulting Hamiltonian is then scaled to fit within the specified energy range, optimizing the algorithm’s performance. Beyond reducing the number of qubits needed for computation, this method concentrates processing power on the subspace containing the reference state, the initial quantum state used as a starting point for the algorithm. “This offers not only the reduction of the qubit count, but also focusing on the subspace that the reference state belongs to,” the authors write, highlighting the dual benefit of their approach. The experiments utilized electronic structure Hamiltonians, specifically those describing Fermi-Hubbard models, commonly used to simulate materials with strong electron interactions, with a ratio of onsite repulsion to hopping energy set at 10, placing the system firmly in a strongly correlated regime. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal. Source: https://iopscience.iop.org/article/10.1088/2058-9565/aea127 More like thisQuantum Research NewsResearchers at Nanoarchitectonics Center Guide Quantum Vortices with Atomic RailsQuantum Research NewsRCQI and CyberSecurityHubCZ co-hosted quantum CEQIP 2026Quantum AlgorithmsResearchers Prove Additivity Links Rényi Divergences to a Minimum FormQuantum AlgorithmsHarvard University Builds Atom Array Control at 84 MFPSStay currentSee today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals. Tags: