Filtered Quantum Phase Estimation

arXiv:2510.04294 · quant-ph, physics.comp-ph · Submitted 2025-10-05 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Filtered Quantum Phase Estimation".

Mira: Accurate state preparation remains a critical bottleneck in many quantum algorithms, particularly those for ground-state energy estimation,

Kai: First, who's behind it and why it matters.

Paper summary: Mira: So, when we look at the title, "Filtered Quantum Phase Estimation," it really encapsulates the core idea that the success of phase estimation depends not just on running more standard repetitions but on intelligently filtering the state preparation process. The authors are showing that this filtering technique is a viable way to tackle the overlap problem inherent in many quantum algorithms, especially those targeting ground-state energies.

Lev: I think the implication for error correction research is that we need to start thinking about state preparation costs not just as a fixed overhead, but as a variable cost that can be mitigated by spectral filtering techniques. If we can tune that trade-off intelligently, it opens up new avenues for designing algorithms that are more efficient on physical hardware.

Kai: From the experimental side, this suggests a path forward where we might start looking at how to implement these filters—whether polynomial or trigonometric—in actual quantum hardware setups to see if those theoretical cost models hold up when we start cooling and measuring.

Mira: Exactly, and the paper's focus on cost awareness means they aren't just proposing a faster algorithm; they are providing a framework for making practical engineering decisions about the implementation trade-offs. It moves beyond just showing that FQPE can be faster than standard QPE in theory.

Lev: And I think for the community, it means we need to be more rigorous about quantifying the impact of state preparation on overall algorithm complexity. It forces a discussion on how much overhead is acceptable when chasing higher precision.

Kai: So, in simple terms, the paper presents a method where we use spectral filtering to get a better initial state overlap, but this comes with known trade-offs regarding how likely we are to succeed and how much time the filter itself takes to run.

Mira: That's the essence of it, and I think that explicit cost awareness is what makes "Filtered Quantum Phase Estimation" significant, as it shows how to navigate those competing demands effectively.

Lev: It gives us a concrete tool for evaluating whether the potential runtime savings from better overlap truly outweigh the added complexity of filtering and postselection.

Kai: So, we've talked about what this paper is all about, from its claims to how it might actually translate into building things on the quantum hardware.

Conclusion: Kai: So, we've looked at how this paper uses spectral filtering to improve state overlap in quantum phase estimation, and now we're getting to the wrap-up where we talk about what this whole "Filtered Quantum Phase Estimation" thing actually means for us.

Mira: I think the title itself is really telling; it points directly to the mechanism—filtering—that they are using to tackle that pesky overlap issue in state preparation, and it makes you wonder how well those assumptions hold up when we move toward real experimental setups.

Lev: From a hardware standpoint, what this means is that we can potentially reduce the number of repetitions needed for phase estimation by significantly improving the initial state quality before even starting the main QPE circuit.

Kai: Exactly, and it suggests that instead of just blindly running more standard QPE circuits until we get a good enough result, we can be smarter about how we prepare our input state first.

Mira: I'm focusing on the authors' framework here; they laid out a very clear cost-aware trade-off between getting a better overlap and the effort required to actually apply that filter function during preparation.

Lev: That trade-off is key, because if the success probability drops too low due to filtering, it could easily negate any speedup we see in the repetition count.

Kai: It sounds like this work isn't just about finding a faster formula; it’s about creating a blueprint for designing algorithms that balance computational cost against state preparation fidelity.

Mira: Precisely; the implication is that for complex simulations where ground states are hard to prepare, this method could be a practical way to make those algorithms feasible on current or near-future quantum devices.

Lev: If we can nail the implementation of those filters—the Gaussian ones or maybe something more adaptive—it gives us a much clearer path for error mitigation strategies specific to state preparation errors.

Kai: So, we've seen the technical details, and now it boils down to what this paper really offers in terms of future direction and real-world impact.

Mira: And that leads us into thinking about how these cost models apply when we start moving beyond simple models toward more realistic physical systems with actual noise profiles.

Gwonhak Lee, Minhyeok Kang, Jungsoo Hong, Stepan Fomichev, Joonsuk Huh

SKKU Advanced Institute of Nanotechnology (SAINT) · Xanadu

quant-ph, physics.comp-ph

Submitted: 2025-10-05

Updated: 2026-06-30

Comments: 44 pages, 14 figures

Journal ref: Quantum Sci. Technol. 11 (2026) 045053

DOI: 10.1088/2058-9565/aea127

Code: https://github.com/snow0369/filtered_quantum_phase_estimation

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 83/100

The gist: Accurate state preparation remains a critical bottleneck in many quantum algorithms, particularly those for ground-state energy estimation, as preparing a state with sufficient overlap with the

Key concepts

Spectral Filtering
This involves applying a bounded function of the Hamiltonian to an initial quantum state to reshape its spectral amplitudes. This process is used to selectively amplify the amplitude corresponding to a desired energy level while suppressing others, aiming for better overlap with that target eigenstate.
Overlap Amplification vs. Success Probability
There is an inherent trade-off: a filter function designed to significantly increase the overlap with the target state must necessarily reduce the probability of successfully preparing that filtered state. This relationship is mathematically bounded, showing you cannot simultaneously maximize both without incurring a penalty.
Filtered Quantum Phase Estimation (FQPE)
This method compares standard QPE against FQPE. While standard QPE requires many repetitions, FQPE reduces the number of repetitions needed. However, each repetition in FQPE now includes the overhead of successfully preparing the filtered state and performing a filtering depth.
Modified Krylov Filters
This is an improvement over standard Krylov filters that introduces a regularization parameter, Lambda. This allows for a tunable trade-off between how sharply the filter converges (convergence sharpness) and how likely it is to succeed (success probability), by making the regularization term dependent on the total cost.

Terminology

Summary

Accurate state preparation remains a critical bottleneck in many quantum algorithms, particularly those for ground-state energy estimation, as preparing a state with sufficient overlap with the desired eigenstate is often challenging. This paper develops a unified cost-aware framework for filtered-state preparation that enhances the overlap of a given input state through spectral filtering, making explicit the trade-off among overlap amplification, preparation success probability, and filter-implementation cost.

The gist

A unified cost-aware framework for filtered-state preparation is developed to enhance the overlap of a given input state through spectral filtering in quantum phase estimation (FQPE), showing that FQPE can reduce total runtime by more than two orders of magnitude in the high-precision regime.

How it works

The framework operates by applying a bounded function of the Hamiltonian, denoted as a filter function, to an initial state to reshape its spectral amplitudes. This filtering process is governed by several key trade-offs:

  1. The success probability of filtered-state preparation is given by the formula:

ϕf⟩ = f(Hˆ)ϕ0⟩∥f(Hˆ)ϕ0⟩∥ = p−1/2 f X d−1 i=0 γif(Ei)Ei⟩.

  1. The resulting target-state overlap is directly related to the initial overlap by:

γf02:= ⟨E0ϕf⟩ 2 = γ0f(E0)2 p−1/2.

  1. A fundamental cost–overlap trade-off is established by combining these, showing that a filter function that significantly amplifies the overlap necessarily exhibits a reduced success probability, bounded by:

pf = γ02γf02f(E0)2 ≤ γ02γf02, where the inequality follows from the boundedness condition f(E0) ≤ 1.

Filtered Quantum Phase Estimation (FQPE)

The generic cost comparison between standard QPE and FQPE is formalized in Theorem 1, which explicitly shows that filtering introduces two competing effects:

  1. The number of QPE repetitions is reduced from O(γ0−2) to O(γf0−2).

  2. Each QPE repetition now requires the successful preparation of the filtered state, incurring a postselection overhead p−1/f and a filtering depth Dsp,f.

The generic cost comparison in terms of expected total circuit depth is stated as:

CQPE = O(γ0−2 log(δ−1) [Dϕ0 + DQPE(ϵ)]

By contrast, filtered QPE estimates E0 with failure probability at most δ using expected total depth C¯FQPE = Oγf0−2 log(δ−1) h p−1/f (Dϕ0 + Dsp,f) + DQPE(ϵ).

Gaussian FQPE with Coarse Spectral Estimates

When a Gaussian filter is used, the framework benefits from coarse estimates of the ground- and first-excited-state energies (E˜0 and E˜1), which are assumed to satisfy E˜0 − E0 ≤ ϵ′∆E0, E˜1 − E1 ≤ ϵ′∆E0. The Gaussian filter is defined by:

g(x) = exp −4(x − µ)2/log ε−1 g ∆2 (Eq. 15).

Theorem 2 provides the cost bound for Gaussian FQPE with these coarse estimates:

C¯FQPE,g(ϵ, δ; ϵ′) = O˜ϵ−1 + γ0−2ϵ−ϵ′∆E/δE−10

Modified Krylov Filters

The paper introduces a modified Krylov-based filter to improve the success-probability/overlap trade-off relative to the standard Krylov construction. This modification involves introducing a regularization parameter Λ, leading to the generalized eigenvalue equation:

[H + Λ(N + 1)I] c = E(N)Λ Sc (Eq. 35).

The choice of the scale for Λ is suggested as Λ = Dsp,f DQPE(ϵ), which makes the regularization term reflect the relative contribution of filtered-state preparation to the total cost in Eq. (10). This modification allows for a tunable and gentle trade-off between convergence sharpness and success probability.

Numerical Comparison of Filter Functions

The numerical experiments on Fermi–Hubbard models demonstrate that Gaussian FQPE can outperform standard QPE when E˜0 and E˜1 are estimated well enough for the filter peak to retain substantial amplitude at E0.

Improvements for AI systems

Based on the provided scientific paper, here are specific improvements that can be made to AI systems and what those improved systems could achieve:

The core contribution of this work is a unified framework for state preparation that explicitly balances the cost of preparing a quantum state (filter implementation cost) against the subsequent quantum computation (QPE runtime). This allows for more efficient and robust eigenvalue estimation in NISQ or near-term fault-tolerant devices.

Here are specific improvements:

  1. Improve Quantum Eigenvalue Estimation Efficiency:

  2. Enable High-Precision Ground State Energy Calculation with Low Initial Overlap:

  3. Develop Adaptive, Cost-Aware State Preparation Routines for Variational Algorithms (VQE):

  4. Enhance Robustness Against Hardware Imperfections in Quantum Circuits:

Specific Improvements and Capabilities:

  1. The system can perform high-precision ground state energy calculations for complex many-body systems (like Fermi-Hubbard models) even when the initial input state has a very poor overlap with the target eigenstate (e.g., overlap as low as 10−4).

  2. The system will achieve a runtime reduction of more than two orders of magnitude compared to standard Quantum Phase Estimation (QPE) in high-precision regimes, specifically when the repetition overhead of standard QPE dominates the total cost.

  3. The system can implement a two-stage algorithm where it first uses standard QPE to obtain coarse spectral estimates, and then uses these estimates to construct an optimized Gaussian filter for a second stage of FQPE, significantly reducing the dependence on the initial overlap by shifting the dominant cost scaling from precision-dependent to gap-scale dependent.

  4. The system can generate a modified Krylov-based filter that offers a tunable trade-off between spectral selectivity (overlap amplification) and postselection success probability, allowing for efficient ground state preparation without needing explicit prior knowledge of the Hamiltonian's spectral structure.

  5. The system can implement filters based on either analytic (polynomial/trigonometric) or data-driven (Krylov subspace) constructions, providing flexibility in circuit depth and spectral selectivity tailored to the specific Hamiltonian being simulated.

  6. The system can incorporate error propagation guarantees using the Davis-Kahan theorem to quantify how perturbations in the implemented quantum filters (due to imperfect qubitization or time evolution approximations) affect the final fidelity of the prepared state, allowing for reliable implementation criteria.

In summary, this research enables a next-generation AI/Quantum simulation system that moves beyond standard QPE by intelligently shaping input states using cost-aware filtering. This leads to a quantum system capable of:

  1. Running high-precision energy estimation on hard problems (low initial overlap).

  2. Achieving massive computational savings by optimizing the trade-off between filter complexity and algorithmic overhead.

  3. Providing a systematic, flexible toolkit for building state preparation routines that are robust to hardware noise and can adapt to the spectral properties of the target problem.

Sources

Related papers