Efficient targeting of arbitrary excited states with quantum inverse power iteration through filtering polynomials

arXiv:2606.28255 · quant-ph, physics.chem-ph · Submitted 2026-06-26 · 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: We're looking at the paper: "Efficient targeting of arbitrary excited states with quantum inverse power iteration through filtering polynomials".

Title and authors: Mira: So we’re focusing on the title and who wrote it first. The paper is "Efficient targeting of arbitrary excited states with quantum inverse power iteration through filtering polynomials." The authors are Srushti Patil and Nina Glaser.

Kai: I see they've put a lot of focus on how this method handles those excited states that are usually buried deep in the energy spectrum. It’s not just about finding *any* state, it’s about finding *arbitrary* ones, which is a big claim for a power iteration method.

Lev: When I look at the authors, they seem to be coming from a solid quantum chemistry background, which is good for understanding the physics involved. But what I wonder is how much of this stability comes from the new technique they propose versus just clever tuning of some parameters.

Kai: Well, they specifically investigate two alternative transformation techniques: Chebyshev decomposition and an eigenstate filtering approach based on QSVT. They found that the filtering approach is substantially more robust because of a symmetry in the applied polynomial, which helps it avoid divergence even when you change the energy shift omega or when you have states that are really close together in energy.

Mira: That robustness is key because those closely spaced spectra are where most of these excited states live, and decomposition-based methods often struggle there. The paper suggests this filtering method efficiently suppresses off-target eigenstates even when they are very close to the target state.

The paper's summary: Kai: So what is the actual core of what they found in this work? Basically, they’re using QIPI based on QSVT to prepare a target excited state by repeatedly applying an approximation of that shifted inverse Hamiltonian. This is different from older methods that used Fourier decompositions and numerical quadrature.

Mira: And the summary highlights that while standard power iteration works for ground states, it becomes computationally prohibitive when you try to access interior eigenstates because the required polynomial degrees just keep getting bigger instead of smaller as you get a better guess for the energy.

Lev: That's the classic problem with direct inversion methods on large quantum systems. It’s not just about needing more math; it's about the sheer cost scaling up too fast for real hardware simulations.

Kai: They introduce two schemes: one based on Chebyshev decomposition, which they found unsuitable because it needs very high polynomial degrees for interior states, and another, the Eigenstate Filtering Polynomial approach. This second scheme constructs a polynomial P that has a value of one at zero and very small values across the rest of the spectrum.

Mira: That filtering polynomial construction is what makes it special. It yields the largest compression of unwanted components among all real polynomials that satisfy that initial condition, which is what gives it its robustness against those closely spaced spectra we talked about earlier.

The paper's improvements: Lev: The results they show for small molecules like H2, LiH, and BeH2 are interesting because they actually show improved convergence and better access to higher excited states compared to other quantum power methods. That’s a tangible win for accessing those higher energy levels.

Kai: And the resource estimates are what really matter for us in this conversation. They provide logical resource estimates in fault-tolerant settings, specifically in terms of T gate counts, which is crucial when we think about running this on actual quantum hardware.

Mira: The paper concludes that QIPI can achieve high target state amplification using only modest polynomial degrees, which they suggest makes it a promising candidate for scalable excited-state preparation in fault-tolerant quantum chemistry applications. It's not demanding the huge polynomial degrees that the decomposition methods required.

Lev: Modest polynomial degrees are what I need to hear when thinking about gate count overhead. If we can keep the degree low, it keeps the T gate count manageable, even if we have to deal with rotation precision issues later on.

Kai: And they do give us a concrete query complexity estimate, which is O of d times k. This simplifies down to something like O of one over delta log one over eta up to some constant factors, which depends on the spectral gap and how accurate you want your target state to be.

Conclusion: Mira: To wrap this up, the paper introduces QIPI as a scalable framework for excited-state preparation that can target any desired eigenpair in a molecular system, not just the low-lying ones. It's really showing that we can tackle arbitrary excited states in this way.

Lev: For real hardware, the iterative structure of QIPI is actually pretty good for noise mitigation because you get some partial mitigation of residual noise through that repeated application, even though the cost still scales with rotation precision and polynomial degree d.

Kai: So what this means for us is that QIPI remains a practical candidate for implementation on future fault-tolerant quantum computers, especially when we consider lower precisions up to one over ten to the power of negative three <ref:2606.28255#pg2>.

Mira: We need to remember that the authors themselves flag a limitation: while it's robust, the per-iteration resource scaling still grows with both the polynomial degree d and one over epsilon, which is our rotation precision.

Lev: So for mid-term fault-tolerant regimes, having a fixed budget on rotation precision seems sufficient to preserve that target state amplification, making QIPI more accessible than we might have thought before.

Kai: It’s a versatile method for reaching any eigenpair in chemistry and many-body physics, and they are making concrete claims about its feasibility in the near future. We'll keep an eye on this paper as it moves toward implementation.

Srushti Patil, Nina Glaser

NNF Quantum Computing Program · Niels Bohr Institute · Department of Chemistry, University of Copenhagen

quant-ph, physics.chem-ph

Submitted: 2026-06-26

Updated: 2026-11-17

Code: https://github.com/LorenzoLaneve/nlft-qsp

Importance score: 83/100

The gist: In this work, we introduce a quantum inverse power iteration (QIPI) algorithm based on the quantum singular value transformation (QSVT) to target arbitrary excited states.

Key concepts

Quantum Inverse Power Iteration (QIPI)
QIPI is an iterative algorithm that uses an approximation of the inverse Hamiltonian to find specific quantum states. It works by repeatedly applying this approximation to a trial state, gradually refining it toward the target state. This iterative process makes it useful for finding states beyond the ground state.
Eigenstate Filtering Polynomial (EF)
This polynomial is constructed specifically to filter out unwanted energy eigenstates during the iteration. It has a property where its values are small on the spectrum of undesired states, allowing the algorithm to efficiently suppress them. This filtering mechanism makes QIPI much more robust than other methods.
Quantum Singular Value Transformation (QSVT)
QSVT is a transformation technique used in this paper to help construct the filtering polynomial. It helps in identifying and manipulating the spectral properties of the Hamiltonian, which is crucial for designing an effective EF that can target arbitrary excited states.

Terminology

Summary

In this work, we introduce a quantum inverse power iteration (QIPI) algorithm based on the quantum singular value transformation (QSVT) to target arbitrary excited states. p. 0 This approach is important because it provides a robust method for targeting arbitrary excited states in fault-tolerant quantum chemistry applications. p. 4

Targeting Excited States

The paper addresses the challenge of accurately determining molecular excited states, which typically require access to a much larger configuration manifold than ground states. Existing methods often struggle with interior or higher-lying eigenstates due to computational cost and instability. p. 1 The authors investigate two alternative transformation techniques, the Chebyshev decomposition of the inverse (Cheb-inv) and an eigenstate filtering (EF) approach based on QSVT. p. 0 They find that EF-based QIPI is substantially more robust than Cheb-inv and other decomposition-based approaches due to the symmetry of the applied filtering polynomial, avoiding divergence with respect to the choice of ω and efficiently suppressing off-target eigenstates even in closely spaced spectra. p. 0

Algorithm Methodology

The QIPI algorithm involves iteratively applying an approximation of the shifted inverse Hamiltonian (H − ωI)−1 to a trial state. The authors present two schemes:

  1. Chebyshev Decomposition-based QIPI, which approximates x−1 as GN(x) using Chebyshev polynomials. p. 7 This method is found to be unsuitable for targeting excited states because it requires very high polynomial degrees to target interior states embedded in dense spectra, and the required degree increases rather than decreases with the quality of the energy guess. p. 9

  2. Eigenstate Filtering Polynomial, which constructs a polynomial P with P(0) = 1 and small values on the remaining spectrum D∆, such that P(H˜) ≈ Pλt. p. 9 This construction is particularly suitable for eigenstate filtering and yields the largest compression of unwanted components among all real polynomials satisfying P(0) = 1 up to degree d. p. 9

Performance and Resource Estimates

Numerical simulations for molecular Hamiltonians of H2, LiH, and BeH2 show improved convergence and enhanced access to higher excited states relative to other quantum power methods. The EFP-based QIPI converges in fewer iterations than existing decomposition-based approaches while maintaining the same query complexity as QSVT. p. 0 The total query complexity is derived as Qtotal = O(dk), which simplifies to O(1/∆ log 1/η) up to constant factors, governed by the spectral gap and target accuracy. p. 7

Feasibility in Fault-Tolerant Settings

When analyzing the per-iteration resource scaling in terms of T-gate count, the cost grows with both the polynomial degree d and 1/ϵ (rotation precision). However, for lower precisions up to 1e−3, a fixed budget on rotation precision is sufficient to preserve target state amplification, making QIPI more accessible for the mid-term FTQC regime. p. 3 The iterative nature of QIPI makes it practically more robust than a single high-order polynomial application because residual noise can be partially mitigated through the iterative structure. p. 9

Query Complexity Comparison

The query complexity of EFP-based QIPI is O(1/γ2∆ log 1/η) without amplitude amplification, matching the near-optimal complexity of QSP-based eigenstate filtering algorithms. With amplitude amplification, the query complexity can be reduced to O(1/γ∆ log 1/η), and a constant ancilla overhead of 3 qubits is required due to the use of QSP and AA. This makes QIPI a promising candidate for scalable excited-state preparation in fault-tolerant quantum chemistry applications. p. 1

Conclusions and Outlook

QIPI provides a scalable and versatile framework for excited-state preparation that can target any desired eigenpair, not just low-lying states. The authors conclude that QIPI remains effective at rotation precisions compatible with fault-tolerant architectures, making it a practical candidate for implementation on future fault-tolerant quantum computers. p. 1 They expect QIPI to be comparatively robust to noise because of its iterative structure. p. 0

Code Availability

All code used to generate the results presented in this work will be made publicly available on github upon publication of the manuscript.

Acknowledgement

This work is supported by the Novo Nordisk Foundation, Grant number NNF22SA0081175, NNF Quantum Computing Programme. Nina Glaser thanks everyone at Riverlane for their generous hospitality and Bjorn Berntson and the Discover team for many stimulating discussions on QSVT and fault-tolerant quantum computing in general during her visit. p. 0 Srushti Patil thanks Maria Gabriela J. p. 0 Oliveira and Marcel David Fabian for valuable discussions on QSVT and excited states. p. 0 Additionally, both authors thank Lorenzo Laneve and Christoph S¨underhauf for making their nlft-qsp package for finding QSP/QSVT angles publicly available. p. 6

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Robust excited state targeting through Eigenstate Filtering Polynomials (EF) allows for suppression of off-target eigenstates with a symmetry of the applied filtering polynomial, avoiding divergence with respect to the choice of omega and efficiently suppressing off-target eigenstates even in closely spaced spectra. This makes the algorithm substantially more robust than decomposition-based approaches.

  2. Scalable excited state preparation in fault-tolerant quantum chemistry applications is achieved by using QIPI with a modest polynomial degrees and achieving high target state amplification with modest polynomial degrees, making it a promising candidate for scalable excited-state preparation in fault-tolerant quantum chemistry applications.

  3. The query complexity of the EFP-based QIPI algorithm is O (1/delta log 1/eta) up to constant factors, which matches the near-optimal query complexity of other methods while targeting arbitrary excited states with comparatively shallow, low-degree iterative circuits and a constant ancilla overhead.

  4. The system can target any desired eigenpair, not just low-lying states, serving as a promising primary algorithmic primitive for applications in quantum chemistry and many-body physics that require access to arbitrary excited states.

Sources

Related papers