Finite-temperature quantum Krylov method from real-time overlaps

arXiv:2604.10543 · quant-ph, cond-mat.str-el · Submitted 2026-04-12 · Read on arXiv

Listen

Transcript

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

Kai: Today's paper: "Finite-temperature quantum Krylov method from real-time overlaps".

Mira: Accurately evaluating finite-temperature properties of quantum many-body systems remains a central challenge,

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

Title and authors: Mira: So, focusing on what they actually achieved, this paper introduces the Finite-temperature quantum Krylov method from real-time overlaps as a way to calculate finite-temperature properties of many-body systems without needing to specify a target temperature on the quantum device.

Kai: That is because it relies solely on the real-time overlap sequence generated by unitary real-time evolution under the Hamiltonian H, which can then be used for classical post-processing.

Lev: It’s essentially using the dynamics of the system itself, captured in those overlaps, rather than having to prepare a thermal state at every temperature point we want to measure.

Kai: They demonstrate that this sequence allows them to obtain thermodynamic quantities like specific heat, magnetic susceptibility, and entropy across a broad temperature range, even down to the T → zero limit <ref:2604.10543#pg1,a broad temperature range, even>.

Mira: This is important because it bypasses the bottleneck of needing thermal state preparation at every target temperature on the quantum hardware for low-temperature calculations.

Lev: That avoids a significant practical limitation where circuit depth and accuracy become very demanding when you try to prepare those thermal states explicitly.

Kai: They also show they can evaluate magnetic susceptibility accurately without needing explicit symmetry-sector decomposition by employing pseudorandom vectors compatible with S z tot conservation.

Mira: That shows the method isn't limited just to energy quantities; it can capture physical observables linked to conserved quantities, which is a strong feature.

Lev: Having that capability means the framework is more versatile than methods that might only be able to calculate spectral information or specific energy values easily.

Kai: The methodology involves an affine transformation of the original Hamiltonian, H˜ = τH + θI, where classical constants map the spectrum into zero π, which helps in recovering the original eigenvalues Ek <ref:2604.10543#pg1,the original Hamiltonian, H˜ = τH + θI>.

Mira: That mapping step is necessary because it sets up a mathematical structure that allows them to utilize those reconstructed energies in their partition function estimates.

Lev: The paper implies that this entire procedure is designed to be applied sector-resolved, meaning they can handle different symmetries within the same general framework.

Kai: They then use this to solve a generalized eigenvalue problem for cos(H˜) using an overlap matrix S(r,q) which has a Toeplitz form.

Mira: That Toeplitz form of the overlap matrix is what allows them to derive approximate eigenvalues and energy estimates E j that they can plug into their partition function formula.

Lev: So the core idea is transforming time evolution information into a set of solvable problems involving eigenvalues derived from these overlaps, which then lead to thermodynamics.

Kai: Finally, they provide a stabilization method by removing small-eigenvalue modes of S and adding corrections to ground state and first excited state energies and their weights.

Mira: That stabilization is what allows them to maintain good reconstruction accuracy even when the input sequence is noisy at the level of sigma = ten−three <ref:2604.10543#pg0>.

Lev: It's a practical engineering detail; if they can stabilize it against that level of noise without resorting to exact diagonalization inputs, that makes it much more viable for real hardware.

The paper's summary: Kai: Looking at the suggested improvements in the paper, one major area is the affine transformation of the Hamiltonian H˜ = τH + θI, where tau and theta are pre-computed classically to map the spectrum into zero π <ref:2604.10543#pg1,are pre-computed classically to map the spectrum>.

Mira: That classical pre-computation is key because it anticipates future applications to larger systems where estimating such a spectral lower bound for a specific sector remains classically feasible even when full finite-temperature simulations are intractable.

Lev: From an error correction view, if we can estimate those lower bounds classically for different sectors, that gives us a roadmap for how much classical overhead we might need before the quantum part becomes the main bottleneck.

Kai: Another improvement is introducing pseudorandom vectors compatible with S z tot conservation to handle magnetic susceptibility evaluation without explicit symmetry-sector decomposition.

Mira: That suggests a more robust approach to calculating observables associated with conserved quantities, which avoids having to perfectly define those sectors on the quantum computer beforehand.

Lev: If we can use pseudorandom vectors instead of relying on exact sector labels, that makes the simulation much less dependent on perfect initial state preparation.

Kai: The method also suggests introducing sophisticated regularization techniques for Krylov subspace methods, like removing small-eigenvalue modes and adaptive truncation thresholds, to maintain stability against sampling errors and noise in the generalized eigenvalue problem formulation.

Mira: Those regularization techniques are necessary because they directly address the inherent instability that can arise from finite sampling or noise in these types of problems.

Lev: So, by focusing on those specific mathematical techniques, the focus shifts from just getting a result to making sure that result is reliable when you're running it on noisy quantum hardware.

Kai: The paper also points toward hybrid workflows where classical post-processing reconstructs complex thermodynamic data from real-time quantum evolution measurements.

Mira: That provides a clear blueprint for how this method fits into existing computational paradigms, allowing classical tools to handle the heavy lifting of data interpretation after the quantum measurement.

Lev: A clear workflow means that we can start designing hybrid setups where the quantum device is used primarily for generating those overlap sequences, and then classical computers do the thermal reconstruction.

The paper's improvements: Kai: So, to wrap up this discussion on "Finite-temperature quantum Krylov method from real-time overlaps," this paper essentially proposes a way to derive finite-temperature properties without needing a fixed target temperature on the quantum device.

Mira: The main implication is that using only the real-time overlap sequence enables thermodynamic quantities like specific heat, magnetic susceptibility, and entropy to be obtained over a broad temperature range with high accuracy.

Lev: This framework seems promising for near-future hardware because it offers a way around the traditional bottlenecks of thermal state preparation at each required temperature.

Kai: We've discussed how they use classical pre-computation and stabilization techniques to handle noise, showing good reconstruction accuracy even under sigma = ten−three <ref:2604.10543#pg0>.

Mira: The paper also provides a clear blueprint for hybrid workflows where classical post-processing reconstructs complex thermodynamic data from real-time quantum evolution measurements.

Lev: It's encouraging to see results that match established benchmarks like those from exact diagonalization and FTLM, which gives us confidence in the method's reliability.

Kai: This work on the Finite-temperature quantum Krylov method from real-time overlaps shows a new way to access finite-temperature information using only real-time overlap sequences.

Mira: It sets a high bar for how we should think about simulating quantum many-body systems at non-trivial temperatures.

Lev: For us, it’s a solid piece of research that points toward what might be achievable with current and near-future quantum hardware if we can manage the resource demands they outlined.

Conclusion: Kai: So, we’ve talked about how they developed the Finite-temperature quantum Krylov method from real-time overlaps to calculate thermodynamics without specifying a target temperature on the quantum device.

Mira: Exactly, and what I find really compelling is how they manage to derive those thermodynamic quantities across such a broad temperature range using only those overlap sequences.

Lev: And from an error correction standpoint, it suggests that if we can reliably generate these real-time overlaps, we might bypass some of the state preparation overhead that plagues other methods for low-temperature calculations.

Kai: Right, and they’ve even shown that this approach works well even when the input sequence is noisy at a level of sigma = ten−three which is important for practical hardware constraints.

Mira: That noise tolerance combined with the classical post-processing reconstruction makes the entire methodology quite robust for studying things like specific heat and magnetic susceptibility.

Lev: If you can manage that level of stability, then running this on real hardware becomes a much more feasible goal because we don't have to worry about perfectly preparing the thermal state first.

Kai: It’s a big step forward for experimentalists because it gives us a new tool to probe these systems dynamically rather than just static measurements at one point in time.

Mira: I think the main implication here is that we gain a powerful, general-purpose method for extracting finite-temperature physics from the time evolution itself.

Lev: It opens up avenues for running error correction protocols that are more flexible with respect to temperature scales on quantum processors.

Kai: So, in summary, the Finite-temperature quantum Krylov method from real-time overlaps gives us a novel path to calculating thermodynamic properties over a wide range using only real-time evolution data.

Mira: It's a testament to how powerful classical post-processing can be when paired with smart quantum measurements.

Lev: It really shows that the bottleneck isn't always the quantum computation itself, but often in how we extract and stabilize those results from the dynamics.

Kai: Speaking of dynamics, I’m eager to hear what other recent papers on arXiv are showing regarding how these methods might integrate with time-dependent simulations for things like spin dynamics.

Department of Physics, Tohoku University

quant-ph, cond-mat.str-el

Submitted: 2026-04-12

Updated: 2026-10-02

Comments: Substantially revised version

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

Importance score: 83/100

The gist: Accurately evaluating finite-temperature properties of quantum many-body systems remains a central challenge, and this work introduces a distinct framework based only on real-time overlap sequences

Key concepts

Real-Time Overlap Sequence: This is the core input from a quantum device, represented as $\langle\phi|e^{-in\tau H|\phi}\rangle$. It is generated by performing unitary real-time evolution under the Hamiltonian $H$ and measuring the overlap between an initial state $|\phi\rangle$ and its evolved state after time $n\tau$. This sequence contains all necessary information for reconstructing thermal properties.
Finite-Temperature Quantum Krylov (FTQK) Method: This method extracts thermodynamic quantities by using the real-time overlap sequence instead of repeated applications of the Hamiltonian. It involves mapping the original Hamiltonian to a transformed one ($\tilde{H}$) to find energy estimates, which are then used in an analogous partition function calculation similar to the Finite-Temperature Lanczos method.
Sector-Resolved Reconstruction: The method performs real-time evolution in the full Hilbert space but restricts sector analysis only to the initial random vectors. This allows for sector-specific transformation parameters ($\tau(q)$ and $\theta(q)$) to be used, enabling a detailed reconstruction of properties across different sectors ($S_{ztot}$ labeled by $q$)
Stabilization Procedure: To handle noise, the method stabilizes results by removing small-eigenvalue modes from the overlap matrix $S$. It also introduces corrections to the approximate ground state energy and first-excited state energy. This stabilization allows for accurate finite-temperature evaluation even when noise is present at a level of $\sigma \sim 10^{-3}$.

Terminology

Accuracy — 19× in this paper · explained in 1 episode
Finite-Temperature Quantum Krylov — 10× in this paper
Benchmark — 6× in this paper · explained in 2 episodes
Hilbert Space — 5× in this paper · explained in 2 episodes
Robustness — 4× in this paper · explained in 5 episodes
Regularization — 3× in this paper · explained in 3 episodes
Frustration — 2× in this paper · explained in 2 episodes
Core — 1× in this paper · explained in 2 episodes
Quantum Circuits — 1× in this paper · explained in 2 episodes
Trace — 1× in this paper · explained in 2 episodes

Summary

Accurately evaluating finite-temperature properties of quantum many-body systems remains a central challenge, and this work introduces a distinct framework based only on real-time overlap sequences that enables thermodynamic quantities to be obtained over a broad temperature range without specifying a target temperature on the quantum device.

The gist: A method is proposed for reconstructing finite-temperature thermodynamic quantities over a broad temperature range, including the T → 0 limit, using only the real-time overlap sequence generated by unitary real-time evolution under the Hamiltonian H.

Motivation and Problem Context

Elucidating finite-temperature properties of quantum many-body systems is important for understanding collective quantum phenomena such as superconducting transitions and quantum spin liquids, as thermodynamic quantities like specific heat and magnetic susceptibility directly reflect these phenomena. However, finite-temperature quantum many-body problems remain notoriously challenging in general, especially when dealing with frustration, high dimensionality, or the sign problem. Existing approaches often require thermal-state preparation at each target temperature on the quantum device, which makes low-temperature calculations demanding in terms of circuit depth and accuracy because thermal-state preparation or temperature-dependent state updates become a practical bottleneck.

The Finite-Temperature Quantum Krylov (FTQK) Method

The proposed method is referred to as the finite-temperature quantum Krylov (FTQK) method. It shares the basic idea of extracting finite-temperature quantities from a Krylov subspace with the Finite-Temperature Lanczos method (FTLM), but fundamentally differs because its input is described not by repeated applications of H, but by overlaps generated by unitary real-time evolution that are naturally accessible on quantum hardware. The only input required from the quantum device is the real-time overlap sequence denoted as:

/n = ⟨ϕe−inτHϕ⟩ (n ∈ Z) (Equation 2). This sequence enables the reconstruction of thermodynamic quantities over a broad temperature range, including the T → 0 limit, through classical post-processing. The method is implemented using an affine transformation to the original Hamiltonian: H˜ = τH + θI, where constants τ and θ are pre-computed classically to map the spectrum of H˜ into [0, π]. This allows for recovery of the original eigenvalues as Ek = (E˜k − θ)/τ. The method is applied separately in each S ztot sector labeled by q, using sector-specific transformation parameters τ(q) and θ(q). Crucially, the implementation retains a full-Hilbert-space realtime-evolution framework while enabling sector-resolved finite-temperature reconstruction. This means the real-time evolution is performed in the full 2N-dimensional Hilbert space, while sector restriction is imposed only on the initial random vectors. The overlap matrix S(r,q) takes a Toeplitz form: S(r,q)nn′ = ⟨ϕ(r,q) nϕ(r,q) n′⟩ = g(r,q)n'−n (Equation 3). The method solves a generalized eigenvalue problem for cos(H˜): F(r,q)u(r,q)j = λ(r,q)j S(r,q)u(r,q)j (Equation 5), yielding approximate eigenvalues λ j and corresponding energy estimates E j. The partition function Z(T) is then estimated analogously to FTLM using these reconstructed energies: Z(T) ≈ M Xsat q=-Msat N(q) st R X R r=1 D Xeff−1 j=0 e−βE(r,q) w(r,q) j (Equation 6). The internal energy, specific heat, magnetic susceptibility, and entropy are evaluated from the reconstructed energy and weight. The method is further stabilized in the presence of noise by removing small-eigenvalue modes of S and introducing further corrections to the approximate groundstate energy, the first-excited-state energy, and their corresponding weights. This stabilization does not rely on direct input from exact diagonalization. The results show that when combined with an appropriate stabilization procedure, the method retains good reconstruction accuracy even under noise at the level of σ ∼ 10−3. In practice, stable finite-temperature evaluation under noise demands increased quantum resources, both in the number of samples and in the amount of real-time overlap data.

Benchmark Results and Accuracy

The proposed method was benchmarked for the one-dimensional spin- 1/2 Heisenberg model with periodic boundary conditions (PBC), focusing on specific heat, magnetic susceptibility, and entropy. In the absence of noise, the results obtained by FTQK agree well with reference results from exact diagonalization (for N=14) and FTLM (for N=24) over a broad temperature range. Notably, the specific heat is reproduced with high accuracy, demonstrating that the method is capable of capturing sensitive finite-temperature thermodynamic quantities.

Improvements for AI systems

Based on the provided scientific paper, here are specific improvements for AI systems and what these improved systems can achieve:

Improve the capability of quantum simulation algorithms to accurately calculate finite-temperature thermodynamic properties of complex many-body quantum systems (like spin models) without requiring repeated thermal state preparations.
Enable the development of new computational frameworks for near-future quantum hardware by leveraging real-time overlap sequences as a primary input, rather than relying on explicit temperature specification or thermal state construction.
Allow AI/Quantum algorithms to reconstruct thermodynamic quantities (specific heat, magnetic susceptibility, entropy) over a broad temperature range, including the low-temperature limit and the high-temperature regime, with high accuracy even when subjected to realistic finite-shot statistical errors (e.g., noise at the level of σ = 10−3).
Facilitate the evaluation of physical observables associated with conserved quantities (like magnetic susceptibility) accurately, even when operating in a full Hilbert space without explicit symmetry sector decomposition on the quantum computer.
Create more robust and stable quantum algorithms for Krylov subspace methods by introducing sophisticated regularization techniques (e.g., removing small-eigenvalue modes, adaptive truncation thresholds, and low-energy corrections) that maintain stability against sampling errors and noise in the generalized eigenvalue problem formulation.
Provide a blueprint for hybrid classical-quantum computation workflows where classical post-processing is used to reconstruct complex thermodynamic data from real-time quantum evolution measurements.
Enable the integration of this method with established classical tensor network approaches (like DMRG, MPS) to approximate real or imaginary-time evolution, thereby extending the applicability of quantum simulation techniques across different computational paradigms.