arxiv: 2604.26047 · v1 · submitted 2026-04-28 · 🪐 quant-ph

Recognition: unknown

Iterative warm-start optimization with quantum imaginary time evolution

Phillip C. Lotshaw , Titus Morris , Stuart Hadfield , Ryan Bennink

Authors on Pith no claims yet

Pith reviewed 2026-05-07 16:14 UTC · model grok-4.3

classification 🪐 quant-ph

keywords quantum optimizationimaginary time evolutionwarm-start methodsMaxCutcombinatorial optimizationnonvariational algorithmshybrid quantum-classical

0 comments

The pith

A nonvariational quantum algorithm iteratively improves combinatorial optimization solutions by evolving a superposition around the current best-known answer via imaginary time evolution.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents an iterative quantum method for approximate combinatorial optimization that begins each step from the best solution found so far rather than a uniform superposition. It constructs a quantum state centered on that solution, applies a nonvariational imaginary time evolution circuit whose form is derived analytically on a classical computer, samples the evolved state, and adopts any improved outcome as the new starting point. Simulations on MaxCut instances for 3-regular graphs of up to 30 vertices, using a total budget of 100 shots, produce median solutions within 95 percent of the global optimum and recover the exact optimum in at least 11 percent of runs. These results exceed those of random sampling and basic classical search procedures. The design therefore offers a concrete way to use limited quantum resources to refine already-good candidate solutions instead of searching from scratch.

Core claim

The central claim is that an iterative procedure—starting with a state superposed around the current best-known solution, driving that state to lower energy with quantum imaginary time evolution whose circuits are fixed by analytic classical equations, sampling to obtain a candidate improvement, and repeating—yields high-quality approximate solutions to combinatorial problems. When applied to MaxCut on 3-regular graphs with at most 30 vertices under a total shot budget of 100, the procedure achieves median approximation ratios of at least 0.95 and recovers the global optimum in 11 percent or more of cases, outperforming random and simplified classical baselines.

What carries the argument

The iterative warm-start loop that applies classically derived, nonvariational quantum imaginary time evolution circuits to a superposition centered on the current best-known solution, thereby driving the state toward lower-energy configurations without variational parameter tuning.

If this is right

The same iterative structure can be applied to other combinatorial optimization problems that admit an energy function and a classical initial solution.
With a fixed total shot budget the method concentrates measurements on refining promising regions instead of exploring the entire space uniformly.
Because the evolution circuits are computed analytically, the quantum device is used only for state preparation and sampling rather than for optimization loops.
Performance improves whenever a reasonably good initial solution is supplied, making the approach naturally compatible with classical heuristics that generate starting points.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Pairing the quantum step with stronger classical warm-start generators could extend useful performance to graphs larger than those simulated here.
The analytic circuit construction may remain tractable for structured problem families even when the underlying graphs grow, provided the imaginary-time equations stay closed-form.
Implementation on noisy intermediate-scale devices would reveal whether the reported shot efficiency survives realistic gate errors and readout noise.

Load-bearing premise

The simulations assume ideal, noise-free quantum hardware and that the analytic classical formulas correctly generate the exact imaginary-time-evolved state from the chosen warm-start superposition.

What would settle it

Running the full iterative procedure on a physical quantum processor for a 20-vertex 3-regular MaxCut instance with exactly 100 shots and measuring whether the median solution quality meets or exceeds 95 percent of the known optimum would directly test the performance claim.

Figures

Figures reproduced from arXiv: 2604.26047 by Phillip C. Lotshaw, Ryan Bennink, Stuart Hadfield, Titus Morris.

**Figure 1.** Figure 1: The iterative quantum optimization algorithm uses a conventional computer (left) to compute a circuit that attempts to decrease the energy of the best-known solution, then sends the circuit to a quantum computer (right) which executes the circuit to update the best-known solution. updated, else the procedure is repeated. The algorithm terminates after a fixed number of iterations, returning the best known … view at source ↗

**Figure 2.** Figure 2: Performance of the protocol for MaxCut on collections random 3-regular graphs. (top) Using 100 total measurements the approach yields median normalized cost values that are close to optimal for N ≤ 30, outperforming random search and a ”classical” version of the protocol that omits the quantum imaginary time evolution. Error bars show quartiles of the distributions. (bottom) The algorithm obtains significa… view at source ↗

**Figure 3.** Figure 3: Performance at N = 22 as the number of iterations Nit and the number of shots per iteration Nshots/it are varied, with a fixed shot budget Nshots = 100. relative to the random search. However, the normalized cost of the final solution is consistently worse than the protocol including U QITE, and decreases more rapidly with system size view at source ↗

read the original abstract

Approximate combinatorial optimization is a promising use case for quantum computers. The quantum optimization algorithms often employ a fixed ansatz that evolves an unbiased initial state towards states with better values of the optimand, then samples the states to determine an approximately optimal solution. However, promising alternative approaches have considered ``warm-start" and sampling-based methods that instead begin from the best known solution, which can be directly optimized with the quantum computer and updated as new information becomes available, potentially outperforming the fixed ans\"atze. Here we use these ideas to design a nonvariational quantum algorithm for combinatorial optimization. At each step the algorithm begins with a state superposed around the best known solution, then drives it to lower energy using quantum imaginary time evolution. These nonvariational, initial-state-dependent circuits are determined using analytic equations that are evaluated using only a conventional computer. After implementing the circuits, the state is sampled, potentially obtaining a new best-known solution to use as the initial state at the next iteration. Using simulations of the algorithm solving MaxCut on 3-regular graphs with 30 or fewer vertices and a shot budget of 100 total shots, the approach obtains median solutions within 95\% of the global optimum and finds optimal solutions in 11\% or more of cases, significantly outperforming random and simplified classical search procedures. We discuss several future directions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper gives a clean iterative warm-start QITE loop for MaxCut whose circuits are fixed by classical analytic equations, with decent small-scale simulation numbers but clear limits on what those numbers actually prove.

read the letter

The main thing to know is that the authors built an iterative algorithm that starts from a classical best-known solution, creates a superposition around it, applies imaginary time evolution through circuits whose parameters come from analytic equations solved on a classical computer, samples the result, and feeds any improvement back into the next round. That structure is the actual novelty here, extending warm-start and QITE ideas into a non-variational iterative form rather than a one-shot or variational procedure.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes an iterative non-variational quantum algorithm for combinatorial optimization that begins with a superposition around the current best-known solution and applies quantum imaginary time evolution (QITE) to drive the state toward lower-energy configurations. The QITE circuits are non-variational and initial-state-dependent, with their parameters computed via analytic equations evaluated entirely on a classical computer; the resulting circuits are then executed on a quantum device and sampled to potentially update the best-known solution for the next iteration. Simulations on MaxCut for 3-regular graphs with at most 30 vertices, using a total shot budget of 100, report median solutions within 95% of the global optimum and recovery of the optimum in at least 11% of instances, outperforming random sampling and simplified classical procedures.

Significance. If the analytic classical derivation of the QITE circuits is shown to faithfully reproduce the target non-unitary evolution, the approach offers a hybrid classical-quantum optimization method that offloads circuit design to classical computation while using quantum sampling for iterative improvement. The reported performance with very low shot counts on small instances is noteworthy and could motivate further development of warm-start non-variational methods as an alternative to fixed-ansatz variational algorithms such as QAOA.

major comments (3)

[Abstract and algorithm description] Abstract and the description of the algorithm: the central performance claims (median 95% of optimum, optimal solutions in ≥11% of cases) rest on the assertion that analytically computed circuits accurately implement imaginary-time evolution starting from the warm-started superposition; without an explicit derivation or verification that the resulting state energies and sampling statistics match those of the imaginary-time Schrödinger equation applied to the initial state, the simulation numbers may reflect a classical surrogate rather than the intended quantum algorithm.
[Simulation results] Simulation results section: all reported metrics are obtained under ideal, noise-free quantum operations; because the algorithm is presented as a quantum method and the circuits are fixed-depth but non-unitary in origin, the absence of any noise model, error analysis, or robustness test means the data do not yet support claims of practical utility on near-term hardware.
[Simulation results] Comparison to baselines: the statement that the method 'significantly outperform[s] random and simplified classical search procedures' is load-bearing for the contribution, yet the manuscript provides no explicit definition or pseudocode for the classical baselines, nor indicates whether they receive equivalent information (initial best-known solution, same total evaluations) as the quantum procedure.

minor comments (2)

[Abstract] The abstract uses the term 'optimand'; replacing it with 'objective function' or 'cost function' would improve readability for a broad audience.
[Simulation results] The manuscript should include a short table or paragraph clarifying the exact shot allocation across iterations and the criterion for accepting a new best-known solution.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback on our manuscript. We address each major comment below with honest responses and indicate planned revisions to strengthen the work.

read point-by-point responses

Referee: [Abstract and algorithm description] Abstract and the description of the algorithm: the central performance claims (median 95% of optimum, optimal solutions in ≥11% of cases) rest on the assertion that analytically computed circuits accurately implement imaginary-time evolution starting from the warm-started superposition; without an explicit derivation or verification that the resulting state energies and sampling statistics match those of the imaginary-time Schrödinger equation applied to the initial state, the simulation numbers may reflect a classical surrogate rather than the intended quantum algorithm.

Authors: We appreciate the referee's emphasis on this foundational point. The circuit parameters are obtained by solving the linear system arising from the imaginary-time Schrödinger equation projected onto the Pauli basis, following the standard QITE derivation (with the warm-start state entering as the initial vector). The manuscript references this procedure but does not reproduce the full algebraic steps. We will add an explicit derivation, including the relevant equations and a small-scale numerical verification that the sampled energies and statistics match direct imaginary-time evolution of the initial state. This revision will confirm that the reported results arise from the intended quantum algorithm. revision: yes
Referee: [Simulation results] Simulation results section: all reported metrics are obtained under ideal, noise-free quantum operations; because the algorithm is presented as a quantum method and the circuits are fixed-depth but non-unitary in origin, the absence of any noise model, error analysis, or robustness test means the data do not yet support claims of practical utility on near-term hardware.

Authors: We agree that ideal simulations alone do not fully demonstrate near-term practicality. The present study isolates the algorithmic behavior under perfect execution to highlight the low-shot performance. In the revision we will insert a dedicated paragraph discussing the expected impact of typical noise channels on the warm-start superposition and the fixed-depth circuits, together with a brief outline of compatible error-mitigation approaches. Comprehensive noisy simulations lie outside the scope of this initial work and are noted as future research, but the added discussion will temper the practical-utility claims appropriately. revision: partial
Referee: [Simulation results] Comparison to baselines: the statement that the method 'significantly outperform[s] random and simplified classical search procedures' is load-bearing for the contribution, yet the manuscript provides no explicit definition or pseudocode for the classical baselines, nor indicates whether they receive equivalent information (initial best-known solution, same total evaluations) as the quantum procedure.

Authors: This criticism is well-founded. The revised manuscript will include explicit definitions, pseudocode, and implementation details for each classical baseline. We will state clearly that every baseline is supplied with the identical initial best-known solution and is limited to the same total number of objective-function evaluations (matching the 100-shot budget). These clarifications will make the comparison transparent and support the performance claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained via analytic circuit construction

full rationale

The paper defines a nonvariational algorithm that constructs initial-state-dependent ITE circuits via analytic equations evaluated classically, then reports direct simulation outcomes on MaxCut instances under ideal noise-free conditions. No step reduces a claimed prediction or performance metric to its own inputs by construction, nor relies on fitted parameters renamed as results, self-citation chains, or smuggled ansatzes. The reported median 95% optimality and 11% optimal-solution rates are straightforward consequences of applying the specified procedure to the chosen graphs and shot budget, with no evident self-definitional loop or statistical forcing.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Review based on abstract only; no explicit free parameters, invented entities, or ad-hoc axioms are stated. The approach assumes standard quantum mechanics for imaginary time evolution and the feasibility of classical analytic circuit computation.

axioms (2)

domain assumption Quantum imaginary time evolution can be used to drive an initial state toward lower-energy configurations corresponding to better optimization solutions
Central mechanism of the algorithm as described in the abstract.
domain assumption Nonvariational circuits for imaginary time evolution can be determined analytically on a classical computer from the initial state
Key enabler for avoiding variational optimization.

pith-pipeline@v0.9.0 · 5542 in / 1404 out tokens · 74669 ms · 2026-05-07T16:14:55.155948+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

25 extracted references · 9 canonical work pages · 1 internal anchor

[1]

A quantum adiabatic evolution algorithm applied to random instances of an np- complete problem

Edward Farhi, Jeffrey Goldstone, Sam Gutmann, Joshua Lapan, Andrew Lundgren, and Daniel Preda. A quantum adiabatic evolution algorithm applied to random instances of an np- complete problem. Science, 292(5516):472–475, 2001

2001
[2]

Adiabatic quantum computation

Tameem Albash and Daniel A Lidar. Adiabatic quantum computation. Reviews of Modern Physics, 90(1):015002, 2018

2018
[3]

A Quantum Approximate Optimization Algorithm

Edward Farhi, Jeffrey Goldstone, and Sam Gutmann. A quantum approximate optimization algorithm. arXiv preprint arXiv:1411.4028, 2014

work page internal anchor Pith review arXiv 2014
[4]

A review on quantum approximate optimization algorithm and its variants

Kostas Blekos, Dean Brand, Andrea Ceschini, Chiao-Hui Chou, Rui-Hao Li, Komal Pandya, and Alessandro Summer. A review on quantum approximate optimization algorithm and its variants. Physics Reports, 1068:1–66, 2024

2024
[5]

Challenges and opportunities in quantum optimization

Amira Abbas, Andris Ambainis, Brandon Augustino, Andreas Bärtschi, Harry Buhrman, Carleton Coffrin, Giorgio Cortiana, Vedran Dunjko, Daniel J Egger, Bruce G Elmegreen, et al. Challenges and opportunities in quantum optimization. Nature Reviews Physics, 6(12):718–735, 2024

2024
[6]

Performant near-term quantum combinato- rial optimization

Titus D Morris, Ananth Kaushik, Martin Roetteler, and Phillip C Lotshaw. Performant near-term quantum combinato- rial optimization. arXiv preprint arXiv:2404.16135, 2024

work page arXiv 2024
[7]

Symmetry enhanced vari- ational quantum imaginary time evolution

Xiaoyang Wang, Yahui Chai, Maria Demidik, Xu Feng, Karl Jansen, and Cenk Tüysüz. Symmetry enhanced vari- ational quantum imaginary time evolution. arXiv preprint arXiv:2307.13598, 2023

work page arXiv 2023
[8]

Imaginary hamiltonian variational ansatz for combinatorial optimization problems

Xiaoyang Wang, Yahui Chai, Xu Feng, Yibin Guo, Karl Jansen, and Cenk Tüysüz. Imaginary hamiltonian variational ansatz for combinatorial optimization problems. Physical Review A, 111(3):032612, 2025

2025
[9]

Optimizing qubo on a quan- tum computer by mimicking imaginary time evolution

Yahui Chai and Alice Di Tucci. Optimizing qubo on a quan- tum computer by mimicking imaginary time evolution. arXiv preprint arXiv:2505.22924, 2025

work page arXiv 2025
[10]

Solving maxcut with quantum imaginary time evolution: R

Rizwanul Alam, George Siopsis, Rebekah Herrman, James Ostrowski, Phillip C Lotshaw, and Travis S Humble. Solving maxcut with quantum imaginary time evolution: R. alam et al. Quantum Information Processing, 22(7):281, 2023

2023
[11]

Warm- starting quantum optimization

Daniel J Egger, Jakub Mareček, and Stefan Woerner. Warm- starting quantum optimization. Quantum, 5:479, 2021

2021
[12]

Classically optimal varia- tional quantum algorithms

Jonathan Wurtz and Peter J Love. Classically optimal varia- tional quantum algorithms. IEEE Transactions on Quantum Engineering, 2:1–7, 2021

2021
[13]

Bridging classical and quantum with sdp initial- ized warm-starts for qaoa

Reuben Tate, Majid Farhadi, Creston Herold, Greg Mohler, and Swati Gupta. Bridging classical and quantum with sdp initial- ized warm-starts for qaoa. ACM Transactions on Quantum Computing, 4(2):1–39, 2023

2023
[14]

Warm-started qaoa with custom mixers prov- ably converges and computationally beats goemans-williamson’s max-cut at low circuit depths

Reuben Tate, Jai Moondra, Bryan Gard, Greg Mohler, and Swati Gupta. Warm-started qaoa with custom mixers prov- ably converges and computationally beats goemans-williamson’s max-cut at low circuit depths. Quantum, 7:1121, 2023

2023
[15]

Convergence guarantee for linearly-constrained combinatorial optimization with a quantum alternating operator ansatz

Brayden Goldstein-Gelb and Phillip C Lotshaw. Convergence guarantee for linearly-constrained combinatorial optimization with a quantum alternating operator ansatz. arXiv preprint arXiv:2409.18829, 2024

work page arXiv 2024
[16]

Warm-started qaoa with aligned mixers converges slowly near the poles of the bloch sphere

Reuben Tate and Stephan Eidenbenz. Warm-started qaoa with aligned mixers converges slowly near the poles of the bloch sphere. In International Conference on Current Trends in Theory and Practice of Computer Science, pages 311–323. Springer, 2025

2025
[17]

The better solution probability metric: Optimizing qaoa to outperform its warm-start solution

Sean Feeney, Reuben Tate, and Stephan Eidenbenz. The better solution probability metric: Optimizing qaoa to outperform its warm-start solution. In 2025 International Conference on Quantum Communications, Networking, and Computing (QCNC), pages 450–458. IEEE, 2025

2025
[18]

Available: https://arxiv.org/abs/2510.26859

Miguel Angel Lopez-Ruiz, Emily L Tucker, Emma M Arnold, Evgeny Epifanovsky, Ananth Kaushik, and Martin Roetteler. A non-variational quantum approach to the job shop scheduling problem. arXiv preprint arXiv:2510.26859, 2025

work page arXiv 2025
[19]

Iterative quantum optimisation with a warm-started quantum state

Haomu Yuan, Songqinghao Yang, and Crispin HW Barnes. Iterative quantum optimisation with a warm-started quantum state. arXiv preprint arXiv:2502.09704, 2025

work page arXiv 2025
[20]

Con- strained quantum optimization via iterative warm-start XY- mixers

David Bucher, Maximilian Janetschek, Michael Poppel, Jonas Stein, Claudia Linnhoff-Popien, and Sebastian Feld. Con- strained quantum optimization via iterative warm-start XY- mixers. arXiv preprint arXiv:2604.02083, 2026

work page arXiv 2026
[21]

Quantum-enhanced markov chain monte carlo for combinatorial optimization

Kate V Marshall, Daniel J Egger, Michael Garn, Francesca Schiavello, Sebastian Brandhofer, Christa Zoufal, and Stefan Woerner. Quantum-enhanced markov chain monte carlo for combinatorial optimization. arXiv preprint arXiv:2602.06171, 2026

work page arXiv 2026
[22]

Quantum-enhanced markov chain monte carlo

David Layden, Guglielmo Mazzola, Ryan V Mishmash, Mario Motta, Pawel Wocjan, Jin-Sung Kim, and Sarah Sheldon. Quantum-enhanced markov chain monte carlo. Nature, 619(7969):282–287, 2023

2023
[23]

Improving quantum approximate optimization by noise-directed adaptive remapping

Filip B Maciejewski, Jacob Biamonte, Stuart Hadfield, and Da- vide Venturelli. Improving quantum approximate optimization by noise-directed adaptive remapping. Quantum, 9:1906, 2025

1906
[24]

Determining eigenstates and thermal states on a quantum computer using quantum imaginary time evolution

Mario Motta, Chong Sun, Adrian TK Tan, Matthew J O’Rourke, Erika Ye, Austin J Minnich, Fernando GSL Brandao, and Garnet Kin-Lic Chan. Determining eigenstates and thermal states on a quantum computer using quantum imaginary time evolution. Nature Physics, 16(2):205–210, 2020

2020
[25]

Variational ansatz-based quantum simulation of imaginary time evolution

Sam McArdle, Tyson Jones, Suguru Endo, Ying Li, Simon C Benjamin, and Xiao Yuan. Variational ansatz-based quantum simulation of imaginary time evolution. npj Quantum Informa- tion, 5(1):75, 2019

2019