arxiv: 2605.09909 · v1 · submitted 2026-05-11 · 🪐 quant-ph

Recognition: 2 theorem links

· Lean Theorem

Symmetry-Protected Basin Localization in Variational Quantum Eigensolvers

Yangshuai Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:45 UTC · model grok-4.3

classification 🪐 quant-ph

keywords variational quantum eigensolverbasin localizationsymmetry protectedmolecular geometryinitialization preconditionerSE(3) covariancestrong correlation

0 comments

The pith

A geometry-conditioned preconditioner maps nuclear positions directly to circuit parameters inside the correlated ground-state basin for variational quantum eigensolvers.

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

Variational quantum eigensolvers frequently fail at the outset when strong electron correlation divides the energy landscape into separate basins and the starting state falls into a non-ground-state one. The work introduces a preconditioner that respects the SE(3) covariance of the molecular Hamiltonian to translate nuclear geometry straight into initial circuit parameters lying inside the correlated ground-state basin. Statevector tests on six stretched molecules show this reduces Hartree-Fock initialization errors by 38 to 6250 times, places several species at sub-millihartree accuracy right away, and shifts the task from escaping bad basins to refining correlation inside the correct one. The mapping occurs before any quantum shots are taken, so the quantum loop only needs to improve the energy within the selected basin.

Core claim

The central claim is that a geometry-conditioned preconditioner P_eq, which maps nuclear geometry R directly to circuit parameters θ0 while enforcing SE(3) covariance of the Hamiltonian, places the initial state inside the correlated ground-state basin for arbitrary ansatzes.

What carries the argument

The geometry-conditioned preconditioner P_eq that uses SE(3) covariance of the molecular Hamiltonian to produce initial circuit parameters from nuclear geometry.

If this is right

Hartree-Fock initialization errors drop by factors between 38 and 6250 times across the six stretched molecules examined.
Sub-millihartree initialization accuracy is reached immediately for CO, LiH, and H8.
N2, H2O, and BeH2 are placed inside the millihartree-scale correlated basin without further adjustment.
Unit success probability is achieved on disordered H10 chains at fixed optimization budget when the mapping is combined with stochastic escape.

Where Pith is reading between the lines

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

The same symmetry constraint may allow similar direct mappings for ansatzes not tested in the benchmarks, reducing the need for hand-tuned initial parameters on larger systems.
Basin localization performed classically before the quantum loop could complement other strategies for avoiding local minima in variational quantum algorithms.
If the mapping generalizes, it would convert the dominant cost in VQE runs from repeated basin-hopping attempts into a single classical preprocessing step.

Load-bearing premise

The SE(3) covariance of the molecular Hamiltonian supplies enough structure to construct a direct, parameter-light mapping from geometry to circuit parameters that lands inside the correlated ground-state basin for any ansatz and molecular size.

What would settle it

Running the preconditioner on a new stretched molecule outside the six tested cases and finding that the resulting energy lies more than a few millihartree above the true ground-state energy would show the basin localization does not hold.

Figures

Figures reproduced from arXiv: 2605.09909 by Yangshuai Wang.

**Figure 1.** Figure 1: resolves the landscape topology: equivariant pre1 0 1 2 3 4 Principal Component 1 (u) 0 5 10 15 20 25 P r i n c i p a l C o m p o n e n t 2 ( v ) N2 (R = 2.5Å) VQE Landscape FCI min. Equivariant init. HF init. 10 2 10 1 10 0 Energy Error relative to Min (Ha) FIG. 1. Basin topology controls VQE trajectories in the strongly correlated regime. Two-parameter energy landscape of stretched N2 at R = 2.5 ˚A with… view at source ↗

**Figure 2.** Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

read the original abstract

Variational quantum eigensolvers fail before optimization begins when strong correlation splits the molecular energy landscape into competing basins and the initial state selects a non-ground-state basin. We introduce a geometry-conditioned preconditioner $\mathcal{P}_{\mathrm{eq}}:\mathbf{R}\mapsto\boldsymbol{\theta}_0$ constrained by the $SE(3)$ covariance of the molecular Hamiltonian, so that nuclear geometry is mapped directly into circuit parameters in the correlated ground-state basin. This basin localization changes the relevant gradient statistics from concentration controlled to curvature controlled. In statevector benchmarks on six stretched molecules, $\mathcal{P}_{\mathrm{eq}}$ reduces Hartree--Fock initialization errors by factors of $38\times$--$6250\times$, reaches sub-mHa initialization in CO, LiH, and H$_8$, and places N$_2$, H$_2$O, and BeH$_2$ in the mHa-scale correlated basin. In disordered H$_{10}$ chains, equivariant basin targeting and stochastic escape reach unit success probability at fixed optimization budget. The procedure performs basin selection before the shot-limited quantum loop; the quantum circuit then refines correlation inside the selected basin.

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

Symmetry-based map from geometry to VQE parameters claims large basin-selection gains on small molecules but stays too high-level to verify.

read the letter

The paper introduces a preconditioner that maps nuclear geometry directly to initial circuit parameters using SE(3) covariance of the Hamiltonian. The goal is to land inside the correlated ground-state basin before any quantum optimization starts, which is a known pain point when Hartree-Fock initialization fails on stretched or strongly correlated molecules. If the numbers are real, this could cut wasted quantum effort by orders of magnitude on initialization alone. The statevector benchmarks on six small stretched molecules report error reductions of 38× to 6250× relative to Hartree-Fock, with sub-mHa starts on CO, LiH, and H8 and mHa-scale placement on the others. The disordered H10 result with unit success probability under fixed budget is the other concrete data point. What the work does cleanly is frame the symmetry constraint as a classical pre-step that changes the optimization statistics from concentration to curvature. That separation is useful even if the map itself needs more scrutiny. The soft spots are the missing pieces that keep the claims from being checkable. No error bars appear, no description is given of how the equivariant map is built or whether it contains fitted constants, and there are no head-to-head comparisons against other symmetry-aware or orbital-based initializers. The argument also gives no general reason why the same map would continue to select the right basin once the ansatz changes or the molecule grows beyond the tested set, which matches the stress-test concern. This is for groups already running VQE on molecular Hamiltonians and looking for better classical pre-processing. A reader who works on practical initialization or equivariant methods will find the idea worth discussing even before the details are filled in. It deserves a serious referee because the problem is real, the symmetry idea is straightforward, and the reported gains are large enough to justify checking the implementation and scope.

Referee Report

3 major / 1 minor

Summary. The paper introduces a geometry-conditioned preconditioner P_eq : R → θ0 for VQE that imposes SE(3) covariance of the molecular Hamiltonian to map nuclear coordinates directly to initial circuit parameters lying inside the correlated ground-state basin. This is claimed to convert the optimization from concentration-controlled to curvature-controlled regimes. Statevector benchmarks on six stretched molecules (CO, LiH, H8, N2, H2O, BeH2) report Hartree-Fock error reductions by 38×–6250×, sub-mHa initialization for three species, and mHa-scale placement for the others; additional tests on disordered H10 chains show unit success probability when combined with stochastic escape.

Significance. If the central construction generalizes, the symmetry-constrained initialization could address a persistent failure mode in VQE for strongly correlated molecules by achieving basin selection classically before the quantum loop begins, without extra quantum resources. The reported error reductions on the tested stretched systems are substantial and the use of external symmetry to reduce free parameters in θ0 is a clear conceptual strength. However, the absence of a general argument or ablations on ansatz dependence and system size limits the assessed significance to the specific cases examined.

major comments (3)

[§4] §4 (statevector benchmarks): the central performance claims of 38×–6250× error reduction and sub-mHa/mHa-scale initialization are stated without error bars, without description of how the preconditioner parameters are obtained, and without comparison to other symmetry-aware or geometry-based initializers, so the quantitative results cannot be verified from the text.
[§3] §3 (construction of P_eq): the claim that SE(3) covariance alone produces a direct mapping into the correlated basin for arbitrary ansatzes is not supported by tests that vary entanglement depth, orbital ordering, or ansatz family; the basin location is potentially ansatz-dependent, so the symmetry constraint does not automatically guarantee the reported initialization quality beyond the specific circuits and molecules tested.
[§3 and §4] §3 and §4: no general proof or scaling argument is given that the equivariant embedding continues to select the correct basin when the molecule size increases beyond the six tested species or when the nuclear geometry deviates further from equilibrium, which is load-bearing for the assertion that basin localization occurs before the quantum optimization loop.

minor comments (1)

The notation and explicit functional form of the mapping P_eq : R → θ0 would benefit from a dedicated equation or pseudocode block to clarify how the SE(3) covariance is enforced in practice.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their careful reading and constructive comments on our manuscript. We address each major comment point by point below, with revisions planned where the concerns are valid and can be addressed through clarification or added discussion.

read point-by-point responses

Referee: [§4] §4 (statevector benchmarks): the central performance claims of 38×–6250× error reduction and sub-mHa/mHa-scale initialization are stated without error bars, without description of how the preconditioner parameters are obtained, and without comparison to other symmetry-aware or geometry-based initializers, so the quantitative results cannot be verified from the text.

Authors: We agree that the benchmark presentation requires improvement for verifiability. In the revised manuscript we will add error bars derived from repeated optimizations or statistical sampling to all reported error reductions and initialization errors. The construction of the preconditioner parameters is specified in Section 3 via the SE(3)-equivariant embedding; we will insert explicit algorithmic steps and a pseudocode listing to make the mapping procedure fully reproducible from the text. While the original submission did not contain side-by-side numerical benchmarks against other geometry-conditioned initializers, we will add a concise conceptual comparison in the discussion section that situates our symmetry-constrained approach relative to Hartree-Fock and other symmetry-adapted starting points. revision: partial
Referee: [§3] §3 (construction of P_eq): the claim that SE(3) covariance alone produces a direct mapping into the correlated basin for arbitrary ansatzes is not supported by tests that vary entanglement depth, orbital ordering, or ansatz family; the basin location is potentially ansatz-dependent, so the symmetry constraint does not automatically guarantee the reported initialization quality beyond the specific circuits and molecules tested.

Authors: The referee is correct that basin membership can depend on ansatz details. Our manuscript does not assert that SE(3) covariance alone guarantees the correlated basin for arbitrary circuits; the covariance constraint is used to reduce the free parameters in θ₀ while preserving the symmetry of the Hamiltonian. We will revise Section 3 to state this limitation explicitly and add a short discussion of how changes in entanglement depth or orbital ordering could affect basin placement. These points will also be flagged as directions for future empirical investigation. revision: yes
Referee: [§3 and §4] §3 and §4: no general proof or scaling argument is given that the equivariant embedding continues to select the correct basin when the molecule size increases beyond the six tested species or when the nuclear geometry deviates further from equilibrium, which is load-bearing for the assertion that basin localization occurs before the quantum optimization loop.

Authors: We acknowledge that the manuscript contains no general proof or scaling analysis. All claims rest on the empirical results for the six molecules and the disordered H₁₀ chains. In the revision we will rephrase Sections 3 and 4 to make clear that basin localization is demonstrated for the tested systems, and we will insert a dedicated limitations paragraph that discusses the absence of a scaling argument and the need for further work on larger or more strongly distorted geometries. revision: yes

standing simulated objections not resolved

Absence of a general mathematical proof or scaling argument establishing basin selection for arbitrary ansatzes, molecule sizes, and nuclear geometries beyond the six tested species

Circularity Check

0 steps flagged

No circularity: symmetry constraint is external and basin claim is benchmarked

full rationale

The derivation introduces P_eq by imposing the external SE(3) covariance of the Hamiltonian on the geometry-to-parameter map. This is a symmetry principle independent of the target basin or the benchmark data. The claim that the resulting θ0 lands inside the correlated ground-state basin is supported by explicit statevector error-reduction numbers on six specific molecules rather than being true by definition or by fitting to those same results. No self-citations, fitted parameters renamed as predictions, or self-definitional equations appear in the construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract-only review prevents exhaustive extraction; the central claim rests on the existence of an SE(3)-equivariant mapping whose explicit functional form and any internal constants are not stated.

axioms (1)

domain assumption The molecular Hamiltonian is SE(3) covariant
Invoked to justify the geometry-conditioned preconditioner construction.

invented entities (1)

P_eq preconditioner no independent evidence
purpose: Maps nuclear geometry to circuit parameters inside the correlated ground-state basin
New object introduced to perform classical basin selection

pith-pipeline@v0.9.0 · 5493 in / 1441 out tokens · 33825 ms · 2026-05-12T04:45:43.032267+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We introduce a geometry-conditioned preconditioner P_eq : R ↦ θ0 constrained by the SE(3) covariance of the molecular Hamiltonian, so that nuclear geometry is mapped directly into circuit parameters in the correlated ground-state basin.
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

This basin localization changes the relevant gradient statistics from concentration controlled to curvature controlled.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

[1]

even/odd

For local Hamiltonians and local generators, tr( ˆG2 k) grows at most polynomially inNwhiledgrows exponentially, so VarHaar[∂kE] =O(poly(N)/d): an exponential suppression arising from the delocalized support of the initialization measure over the unitary manifold. Basin-localized regime: Theorem S1 Assumption A (strongly convex basin).There exists a neigh...

work page
[2]

Peruzzo, J

A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O’Brien, Nat. Commun.5, 4213 (2014)

work page 2014
[3]

J. R. McClean, J. Romero, R. Babbush, and A. Aspuru-Guzik, New J. Phys.18, 023023 (2016)

work page 2016
[4]

Kandala, A

A. Kandala, A. Mezzacapo, K. Temme, M. Takita, M. Brink, J. M. Chow, and J. M. Gambetta, Nature549, 242 (2017)

work page 2017
[5]

Y. Cao, J. Romero, J. P. Olson, M. Degroote, P. D. Johnson, M. Kieferov´ a, I. D. Kivlichan, T. Menke, B. Peropadre, N. P. D. Sawaya, S. Sim, L. Veis, and A. Aspuru-Guzik, Chem. Rev.119, 10856 (2019)

work page 2019
[6]

Tilly, H

J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth,et al., Phys. Rep.986, 1 (2022)

work page 2022
[7]

D. Wang, O. Higgott, and S. Brierley, Phys. Rev. Lett.122, 140504 (2019)

work page 2019
[8]

H. R. Grimsley, S. E. Economou, E. Barnes, and N. J. Mayhall, Nat. Commun.10, 3007 (2019)

work page 2019
[9]

Cerezo, A

M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio,et al., Nat. Rev. Phys.3, 625 (2021)

work page 2021
[10]

Bittel and M

L. Bittel and M. Kliesch, Phys. Rev. Lett.127, 120502 (2021)

work page 2021
[11]

Z.-C. Yang, A. Rahmani, A. Shabani, H. Neven, and C. Chamon, Physical Review X7, 021027 (2017)

work page 2017
[12]

S. Wang, E. Fontana, M. Cerezo, K. Sharma, A. Sone, L. Cincio, and P. J. Coles, Nat. Commun.12, 6961 (2021)

work page 2021
[13]

Song, C.-M

X.-Y. Song, C.-M. Jian, and L. Balents, Phys. Rev. Lett.119, 216601 (2017)

work page 2017
[14]

Dagotto, Science309, 257 (2005)

E. Dagotto, Science309, 257 (2005)

work page 2005
[15]

J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, and H. Neven, Nat. Commun.9, 4812 (2018)

work page 2018
[16]

Larocca, S

M. Larocca, S. Thanasilp, S. Wang, K. Sharma, J. Biamonte, P. J. Coles, L. Cincio, J. R. McClean, Z. Holmes, and M. Cerezo, Nature Reviews Physics7, 174 (2025)

work page 2025
[17]

C. A. Coulson and I. Fischer, Philos. Mag.40, 386 (1949)

work page 1949
[18]

D. I. Lyakh, M. Musia l, V. F. Lotrich, and R. J. Bartlett, Chem. Rev.112, 182 (2012)

work page 2012
[19]

Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds

N. Thomas, T. Smidt, S. Kearnes, L. Yang, L. Li, K. Kohlhoff, and P. Riley, Tensor field networks: Rotation- and translation- equivariant neural networks for 3d point clouds (2018), arXiv:1802.08219 [cs.LG]

work page Pith review arXiv 2018
[20]

Batzner, A

S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt, and B. Kozinsky, Nature Communications13, 2453 (2022)

work page 2022
[21]

K. T. Sch¨ utt, O. T. Unke, and M. Gastegger, inProceedings of the 38th International Conference on Machine Learning, Proceedings of Machine Learning Research, Vol. 139 (2021) pp. 9377–9388

work page 2021
[22]

F. B. Fuchs, D. E. Worrall, V. Fischer, and M. Welling, inAdvances in Neural Information Processing Systems, Vol. 33 (2020) pp. 1970–1981

work page 2020
[23]

Geiger and T

M. Geiger and T. Smidt, e3nn: Euclidean neural networks (2022), arXiv:2207.09453 [cs.LG]

work page arXiv 2022
[24]

Batatia, D

I. Batatia, D. P. Kovacs, G. N. C. Simm, C. Ortner, and G. Cs´ anyi, Advances in Neural Information Processing Systems (NeurIPS) 35, 11423 (2022)

work page 2022
[25]

Y. Wang, D. Clark, S. Das, Z. Zhu, D. Massatt, V. Gavini, M. Luskin, and C. Ortner, Mach. Learn.: Sci. Technol.6, 045040 (2025)

work page 2025
[26]

Q. Sun, T. C. Berkelbach, N. S. Blunt, G. H. Booth, S. Guo, Z. Li, J. Liu, J. D. McClain, E. R. Sayfutyarova, S. Sharma, S. Wouters, and G. K.-L. Chan, WIREs Comput. Mol. Sci.8, e1340 (2018)

work page 2018
[27]

R. H. Byrd, P. Lu, J. Nocedal, and C. Zhu, SIAM J. Sci. Comput.16, 1190 (1995)

work page 1995
[28]

Ortner and Y

C. Ortner and Y. Wang, Multiscale Model. Simul.21, 1053 (2023)

work page 2023
[29]

Y. Wang, S. Patel, and C. Ortner, Comput. Methods Appl. Mech. Eng.422, 116831 (2024)

work page 2024
[30]

X. Liu, K. Zeng, Z. Luo, Y. Wang, T. Zhao, and Z. Xu, J. Appl. Phys.139, 041101 (2026)

work page 2026
[31]

H. Chen, C. Ortner, and Y. Wang, Multiscale Model. Simul.20, 1490 (2022)

work page 2022
[32]

J. T. Seeley, M. J. Richard, and P. J. Love, J. Chem. Phys.137, 224109 (2012)

work page 2012
[33]

J. C. Spall, IEEE Trans. Autom. Control37, 332 (1992)

work page 1992
[34]

C. H. Ho, C. Ortner, and Y. Wang, npj Comput. Mater. 10.1038/s41524-026-02080-3 (2026), article in press

work page doi:10.1038/s41524-026-02080-3 2026
[35]

Loshchilov and F

I. Loshchilov and F. Hutter, inInternational Conference on Learning Representations(2019)

work page 2019