Quantum dynamics in frustrated Ising fullerenes Authors: Alejandro Lopez-Bezanilla, William Bernoudy, Kelly Boothby, Jack Raymond, Alberto Nocera, Andrew D. King Published: 2026 ABSTRACT The complex energy landscapes exhibited by frustrated magnetic systems undergoing quantum fluctuations are a challenge to accurately simulate, and thus of great interest for testing diverse qubit platforms in the field of quantum simulation. This study experimentally demonstrates quantum fluctuations lifting the degenerate ground-state manifold of classical magnetic configurations in fullerene Ising models with resonating dimers. The interplay between degeneracy and quantum fluctuations makes these boundary-free models well suited as a diagnostic test case for quantum simulators. Indeed, we observe significant performance improvement across generations of superconducting quantum annealers, showing the potential of highly symmetric, frustrated systems for assessing the precision of quantum-simulation technologies.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1140/epjqt/s40507-026-00536-5. Magnetic frustration arises from the inability of magnetic moments to simultaneously satisfy all energetic terms; this can result in highly degenerate ground-state manifolds with extensive ground-state entropy [1, 2]. Quantum fluctuations partially resolve this degeneracy by allowing the system to explore superpositions of classically degenerate states through mechanisms like tunneling [3]. Quantum terms, such as a transverse field applied to an Ising model, mix these configurations and effectively select states that minimize the overall energy [4]. This fluctuation-induced selection mechanism can, in extended systems, lead to spontaneous symmetry breaking and the emergence of long-range order—phenomena broadly associated with order-by-disorder [5–9]. However, in finite systems such as those studied here, exact symmetry breaking cannot occur. Instead, quantum fluctuations lift degeneracy through perturbative corrections that favor specific coherent configurations, which may exhibit signatures of emergent order or preserve all symmetries, as in quantum spin liquid states [10, 11]. Geometrically frustrated systems, with their rich array of quantum phases and selection mechanisms, are therefore especially appealing for studying quantum dynamics and for evaluating the capabilities of emerging quantum simulation platforms. Simulating frustrated magnetic systems requires precise control over the Hamiltonian, as phenomena such as order-by-disorder depend on the accurate realization of the degenerate classical ground-state manifold to serve as the foundation for quantum effects. In recent years, quantum experiments have been conducted in Ising-like systems manufactured from controllable building blocks, including superconducting qubits [12–16], neutral atoms [17–19], and trapped ions [20]. Together, these platforms represent a complementary toolkit, offering diverse approaches for simulating quantum effects. To facilitate direct comparisons across quantum-simulation platforms, this paper proposes a series of boundary-free frustrated lattices and demonstrates their use with quantum annealers based on superconducting qubits. For a series of boundary-free frustrated magnetic lattices, embedded in D-Wave quantum annealers, this paper demonstrates the partitioning by quantum fluctuations of the ground-state manifold of classical magnetic configurations. By facilitating quantum tunneling and local resonance processes within magnetic dimers, and within the coherent-time window where thermal relaxation is not yet dominant, we observe that energetically equivalent classical spin configurations paired into singlet bonds experience energy reduction depending on their local environment. These systems offer a hardware-efficient and versatile cross-platform benchmark that can potentially be evaluated in superconducting quantum annealers, neutral-atom arrays [19, 21], and trapped-ion systems [20, 22, 23]. All systems have uniform-magnitude couplings Jij=±1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$J_{ij} = \pm 1$\end{document} on the bonds ij of a three-regular fullerene graph (Fig. 1); due to the odd connectivity, any two isoenergetic classical states must have Hamming distance ≥2. As such, a perturbative transverse field lifts the classical ground-state degeneracy with a second-order tunneling term arising from resonating bonds. Accurately reproducing the density of such bonds in out-of-equilibrium quantum dynamics is a highly sensitive task for quantum simulation, which we demonstrate here. Figure 1Geometrically-frustrated Ising boundary-free systems. Representation of the antiferromagnetic 20-node dodecahedron, a and 60-node fullerene b, and c. The fullerene’s blue and red edges in c form two symmetry classes that can be assigned ferromagnetic and antiferromagnetic exchange respectively, leading to greater frustration and higher ground-state degeneracy, NGS\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$N_{\mathrm{G}S}$\end{document}, than in the purely antiferromagnetic case b Geometrically-frustrated Ising boundary-free systems. Representation of the antiferromagnetic 20-node dodecahedron, a and 60-node fullerene b, and c. The fullerene’s blue and red edges in c form two symmetry classes that can be assigned ferromagnetic and antiferromagnetic exchange respectively, leading to greater frustration and higher ground-state degeneracy, NGS\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$N_{\mathrm{G}S}$\end{document}, than in the purely antiferromagnetic case b We simulate the lattices displayed in Fig. 1. Nearest-neighbor couplings Jij\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$J_{ij}$\end{document} are defined on a fullerene structure [24–26], a closed mesh in which sites have coordination number three and elementary plaquettes have five or six sites. Specifically we study the dodecahedron (N=20N=20) and Buckminsterfullerene (N=60N=60; buckyball) structures. The dodecahedron (Fig. 1a) consists of 20 qubits interconnected in a pentagonal topology, and exhibits 250 degenerate ground states in a nearest-neighbor all-AFM coupling. The odd cycles in the fullerene graph shown in Fig. 1b ensure magnetic frustration, leading to ∼104 classically degenerate ground states when enforcing an all-AFM coupling between neighboring qubits of the Ising model. In Fig. 1c, two types of couplers are introduced in the system: red couplers connect qubits i and j in AFM interactions with Jij=+1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$J_{ij}=+1$\end{document} while blue couplers establish a ferromagnetic (FM) interaction between two qubits with Jij=−1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$J_{ij}=-1$\end{document}. Red couplers form the five-site cycles, whereas blue couplers interconnect qubits in distinct five-site cycles; each qubit is incident to two red couplers and one blue coupler. Notably, couplers can be classified according to their automorphism class, indicating the symmetry of the system: the graph representing the spins is vertex transitive, meaning that any spin can mapped to any other spin by a graph automorphism. This combination of FM and AFM interactions renders the lattice “fully frustrated”, with both pentagonal and hexagonal plaquettes frustrated, resulting in over a million ground states. The transverse-field Ising model (TFIM) Hamiltonian governing the quantum dynamics of these fullerenes is given by 1H=ΓHD+JHI\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document} $$\begin{aligned} \mathcal {H} &= \Gamma \mathcal {H}_{D} + \mathcal {J}\mathcal {H}_{I} \end{aligned}$$ \end{document}2HD=−∑iσix\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document} $$\begin{aligned} \mathcal {H}_{D} &= -\sum _{i}{\sigma _{i}^{x}} \end{aligned}$$ \end{document}3HI=∑Jijσizσjz,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document} $$\begin{aligned} \mathcal {H}_{I} &= \sum _{< i,j>}{J_{ij}\sigma _{i}^{z}\sigma _{j}^{z}}, \end{aligned}$$ \end{document} where σix\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\sigma _{i}^{x}$\end{document} and σiz\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\sigma _{i}^{z}$\end{document} are Pauli operators; Γ and J\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {J}$\end{document} are the transverse field and the classical Ising energy scale, which respectively weight the driver Hamiltonian HD\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {H}_{D}$\end{document} and the Ising Hamiltonian HI\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {H}_{I}$\end{document}. These systems can be embedded directly as subgraphs of the qubit connectivity graphs of the quantum computers used in this study. The minimum instantaneous spectral gap along the annealing path governs how readily the system can follow the ground state under a finite-rate schedule. For the moderate system sizes considered here, the high symmetry and regular connectivity of the graphs suggest that this gap remains sufficiently large to support coherent evolution over a broad range of annealing times. This is consistent with the observation that the quantum annealer closely tracks the perturbative ground state across that regime [27]. When Γ=0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\Gamma{=}0$\end{document} and J>0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {J}{>}0$\end{document}, these three Ising models have respective classical ground-state degeneracies NGS=250\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$N_{\mathrm{GS}}=250$\end{document}, 16,000\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$16{,}000$\end{document}, and 1,125,000\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$1{,}125{,}000$\end{document}. Upon the addition of a perturbative transverse field, say Γ=ϵ→0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\Gamma{=}\epsilon{\rightarrow}0$\end{document} and J=1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {J}{=}1$\end{document}, quantum fluctuations lift the degeneracy of the classical ground-state manifold, and the quantum ground state becomes a superposition of classical ground states. The amplitudes of this perturbative ground state do not reflect the “fair sampling” of the T=0T=0 Boltzmann distribution [28–31], but rather a distinct and delicate quantum state that we propose here as a reference target for quantum simulation. We can consider a transverse field in the perturbative limit Γ=ϵ→0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\Gamma =\epsilon \rightarrow 0$\end{document}. The ground state of ϵHD+HI\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\epsilon \mathcal {H}_{D}+\mathcal {H}_{I}$\end{document} is a superposition of classical ground states, given by the principal eigenvector of the tunneling matrix among the classical ground states: rows and columns correspond to classical ground states and an entry corresponding to two states differing by a dimer flip is −1, or zero otherwise [32, 33]. We can compare with the uniform symmetric superposition of classical ground states. This has already been done in the case of single-spin tunneling in small Villain models [32]. Any two classical ground states with Hamming distance =2 differ by a “floppy dimer”, meaning that two coupled spins can cotunnel between the two configurations upon the application of quantum fluctuations. A non-floppy dimer and a floppy dimer are depicted in Fig. 2a and b. Figure 2c and d show two distinct classical ground states of a dodecahedron, with floppy dimers indicated as thick gray lines. Figure 2Floppy dimers and resonance in an AFM dodecahedron. Here, all couplers are antiferromagnetic with Jij=1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$J_{ij}=1$\end{document}. A floppy dimer is formed by pairing two spins with opposite orientations (blue and red) in a manner that facilitates consistent observation of the same type of nearest neighbor spins through transverse-field mediated tunneling within the dimer (exchange of spin orientations). A regular dimer in a tunnels between energy levels split by a gap ΔE=4J\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\Delta _{E}=4J$\end{document}, while a “floppy dimer” in b (thick gray line) tunnels between degenerate classical states. Both c and d depict classical ground states, with floppy dimers indicated as thick couplings. Their overlaps with ψϵ\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\psi _{\epsilon}$\end{document}, , are determined by the structure of how floppy dimers connect the classical ground-state manifold. For c, is below average among the 250 classical ground states, whereas the value for d, 0.0063, is above average. These represent two of the five symmetry classes of classical ground states under site relabeling Floppy dimers and resonance in an AFM dodecahedron. Here, all couplers are antiferromagnetic with Jij=1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$J_{ij}=1$\end{document}. A floppy dimer is formed by pairing two spins with opposite orientations (blue and red) in a manner that facilitates consistent observation of the same type of nearest neighbor spins through transverse-field mediated tunneling within the dimer (exchange of spin orientations). A regular dimer in a tunnels between energy levels split by a gap ΔE=4J\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\Delta _{E}=4J$\end{document}, while a “floppy dimer” in b (thick gray line) tunnels between degenerate classical states. Both c and d depict classical ground states, with floppy dimers indicated as thick couplings. Their overlaps with ψϵ\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\psi _{\epsilon}$\end{document}, , are determined by the structure of how floppy dimers connect the classical ground-state manifold. For c, is below average among the 250 classical ground states, whereas the value for d, 0.0063, is above average. These represent two of the five symmetry classes of classical ground states under site relabeling We simulate the nonequilibrium approach toward using quantum processing units (QPUs) from two generations of D-Wave annealing quantum computers: an Advantage™ system and an Advantage2™ prototype; we denote these ADV1 and ADV2 respectively as in Ref. [34]. They realize time-dependent transverse-field Ising Hamiltonians 4H(s)=Γ(s)HD+J(s)HI,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathcal {H}(s) = \Gamma (s)\mathcal {H}_{D} + \mathcal {J}(s)\mathcal {H}_{I}, $$\end{document} where Γ(s)Γ (s) and J(s)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {J}(s)$\end{document} are the transverse and Ising energy scales at normalized time s=t/ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$s=t/t_{a}$\end{document} given by each processor’s annealing schedule (see Fig. S1). In the fast-quench limit ta→0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}\rightarrow 0$\end{document}, the action of the Hamiltonian becomes nil and the output reflects the initial state, which is the ground state of HD\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {H}_{D}$\end{document}: equal superposition of computational basis states. In other words, the final quenched state at t=ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t=t_{a}$\end{document} should be a uniform distribution of all classical states (random, uncorrelated output). Although the Hamiltonian is parametrized by the normalized time s=t/ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$s=t/t_{a}$\end{document}, the total annealing time ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} sets the physical time scale of the evolution. Different values of ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} correspond to different quench rates, and thus affect how closely the system tracks the instantaneous ground state by the end of the anneal (s=1)(s=1). In the slow-quench limit ta→∞\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a} \rightarrow \infty $\end{document}, a closed and coherent quantum system is expected to adiabatically follow the instantaneous ground state of the Hamiltonian, reaching . Experimental systems can reach this state only if they can anneal slowly enough to remain adiabatic while still retaining coherence, which prevents coupling to the environment. (We emphasize that the s-dependent ground state is discontinuous at s=1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$s{=}1$\end{document}, and is the ground state of the system in the limit Γ→0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\Gamma \rightarrow 0$\end{document}; the ideal system spends no time at Γ=0Γ =0 and therefore will not discontinuously jump to the ground state at Γ=0Γ =0, i.e., ). However, due to the limited coherence and precision of experimental platforms, we expect a thermalizing system to explore a broad space of low-energy classical states, deviating from the statistics of toward . This deviation provides a quantitative diagnostic of simulator performance by probing a platform’s ability to maintain fidelity to the target quantum state, which requires precise control and advanced hardware. While fidelity is used here as a validation tool, the broader assessment strategy does not rely on full classical simulation. Instead, we focus on the emergence of a quantum-selected ground state from a degenerate classical manifold, driven by perturbative transverse-field effects. The structure of this splitting, accessible through analytic perturbation theory, provides a physically interpretable fingerprint of quantum coherence. As system sizes grow beyond classical reach, this perturbative structure remains a viable reference against which quantum devices can be meaningfully assessed. Along with quantum annealing (QA), we use exact diagonalization (ED) and matrix-product state (MPS) methods to simulate the model dynamics of the time-dependent QA Hamiltonian for varying ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document}. Although the QPUs’ fastest accessible ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} is 5ns\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$5~\mbox{ns}$\end{document}, we can simulate faster quenches by reducing the magnitude of Jij\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$J_{ij}$\end{document}. We plot all data using a calibrated equivalent time, assuming |Jij|=1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$|J_{ij}|=1$\end{document}: all such ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} are given relative to the ADV1 annealing schedule as in Ref. [34] with physical annealing times for ADV2 roughly 1.75 times faster; details of the fidelity-based time calibration are given in the Appendix. MPS methods are as in Ref. [34]; QA methods use public API settings available to all users [35]. Figure 3 shows the residual energy density, defined by: 5δE=1−〈HI〉E0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \delta _{E} = 1- \frac{\langle \mathcal {H}_{I}\rangle}{E_{0}} $$\end{document} where E0<0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$E_{0} < 0$\end{document} is the ground-state energy of the classical Ising Hamiltonian HI\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {H}_{I}$\end{document} at the end of the anneal. These observables are in good agreement between classical (ED and MPS) and quantum (QPU) experiments for short annealing times, but effects of QPU decoherence are apparent for ta>80ns\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a} > 80~\mbox{ns}$\end{document}, consistent with similar observations in analog QA [36], digitized QA [15] and digital-analog QA [16]. ADV2, which has lower noise and a higher energy scale than ADV1, achieves lower energies both before and after the onset of thermal excitations. Figure 3Residual energy density. As the system approaches the classical ground-state manifold, residual energy density δE\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\delta _{E}$\end{document} decreases sharply from 1 as ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} increases from the fast-anneal limit. Classical simulations values are shown in black: (+: exact diagonalization for N=20N=20; ×: MPS for N=60N=60). They show good agreement with observed QA dynamics up to the point where environmental effects arise. The lower-noise ADV2 QPU drives to significantly lower energies than ADV1. For longer anneals (ta>80ns\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}>80~\mbox{ns}$\end{document}), large step sizes lead to inconsistent energies in MPS Residual energy density. As the system approaches the classical ground-state manifold, residual energy density δE\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\delta _{E}$\end{document} decreases sharply from 1 as ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} increases from the fast-anneal limit. Classical simulations values are shown in black: (+: exact diagonalization for N=20N=20; ×: MPS for N=60N=60). They show good agreement with observed QA dynamics up to the point where environmental effects arise. The lower-noise ADV2 QPU drives to significantly lower energies than ADV1. For longer anneals (ta>80ns\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}>80~\mbox{ns}$\end{document}), large step sizes lead to inconsistent energies in MPS Figure 4a shows D, the average number of floppy dimers in the classical output states, which can be defined as an operator 6D=∑(1−Ji,i′Ji,i″σ′zσ″z)(1−Jj,j′Jj,j″σ′zσ″z)/4,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ D = \sum _{< i,j>}(1-J_{i,i'}J_{i,i''}\sigma _{i'}^{z}\sigma _{i''}^{z})(1-J_{j,j'}J_{j,j''} \sigma _{j'}^{z}\sigma _{j''}^{z})/4, $$\end{document} where for a bond ij, the other two neighbors of i are i′i' and i″i'', and the other two neighbors of j are j′j' and j″j''. Figure 4Floppy dimers and fidelity. a, In an accurate simulation, the average number of floppy dimers, D, should reflect the perturbative ground state in the large-ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} limit. In a thermalized quantum simulation, we expect the statistics of sampling from a uniform superposition over classical ground states . For fast anneals, we observe nonmonotonicity in D arising from competition between excited and ground states; this is reflected in both quantum and classical experiments. b, The binned fidelity F′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {F}'$\end{document}, calculated here over 105 samples, quantifies adherence to the perturbative ground state , and clearly differentiates simulation quality between ADV1 and ADV2. Infidelity for 105 fair samples (95% confidence interval) is below the axis limits. In most cases, the resulting uncertainties are smaller than the marker size, which is why they are not visually apparent in the figure Floppy dimers and fidelity. a, In an accurate simulation, the average number of floppy dimers, D, should reflect the perturbative ground state in the large-ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} limit. In a thermalized quantum simulation, we expect the statistics of sampling from a uniform superposition over classical ground states . For fast anneals, we observe nonmonotonicity in D arising from competition between excited and ground states; this is reflected in both quantum and classical experiments. b, The binned fidelity F′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {F}'$\end{document}, calculated here over 105 samples, quantifies adherence to the perturbative ground state , and clearly differentiates simulation quality between ADV1 and ADV2. Infidelity for 105 fair samples (95% confidence interval) is below the axis limits. In most cases, the resulting uncertainties are smaller than the marker size, which is why they are not visually apparent in the figure For N=20N=20, we see nonmonotonicity in D arising from competition between HD\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {H}_{D}$\end{document} and HI\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {H}_{I}$\end{document}, and this competition is reproduced by both QPUs. Near ta=30ns\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}=30~\mbox{ns}$\end{document}, D plateaus in the MPS data at a value of 5.45 (consistent with ), whereas QPU results tail off, thermalizing in the large-ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} limit to a value of 4.8 (consistent with ).he precise tracking of D by the QPUs in the coherent region of the anneal highlights the utility of frustrated fullerenes as a reference testbed for quantum simulation. In Fig. 4b, we isolate a quality measure: the fidelity with respect to the targeted perturbative ground state . We use the Battachyyara coefficient F=∑ipiqi\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {F}= \sum _{i}\sqrt{p_{i}q_{i}}$\end{document}, where pi\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$p_{i}$\end{document} and qi\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$q_{i}$\end{document} are the probability weights of computational basis state i in the (classical or quantum) simulation and in , respectively. The large entropy of the ground-state manifold leads to sampling error, which we mitigate by binning probabilities based on automorphism orbits, calling the resulting “binned fidelity” F′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {F}'$\end{document}. Still, sampling error remains a significant challenge in estimating F′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal {F}'$\end{document}, so one must be careful to compare fidelities between equal sample sizes—in this case we use 105 samples for all solvers; statistical floors are below the limits of the plots. To effectively distinguish the quality of two simulators, sampling error must be significantly lower than the error of at least one of the simulators. The error bars correspond to 95% confidence intervals, computed using a bootstrap resampling procedure across independent QPU calls, rather than across individual samples within a single call. In Fig. 4b, we see a clear minimum in 1−F′\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$1-\mathcal {F}'$\end{document} for the QPUs as ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} increases; the increase of infidelity after this minimum is a signature of relaxation throughout the classical ground-state manifold. In all cases, error bars are included for all sampled quantities but are typically smaller than the plotting markers and therefore not visually discernible. They represent 95% confidence intervals obtained via bootstrap resampling across independent QPU calls as opposed to resampling individual samples within a single call. We note that in the absence of longitudinal fields, the Hamiltonian preserves global spin-inversion symmetry, resulting in a two-fold degenerate ground state. However, the observables considered in this study—such as the number of floppy dimers and basis state probabilities—are invariant under spin inversion. As such, the symmetric superposition of the degenerate states serves as an appropriate reference for comparison, and no ambiguity arises in defining the reference quantities used to quantify quantum selection in the perturbative regime. As the field of quantum simulation advances, with diverse platforms demonstrating precise control over complex quantum states, evaluating and improving these technologies requires the development of tools that reveal nontrivial quantum behavior and facilitate direct, cross-platform comparisons between qubit modalities. This work introduced frustrated Ising fullerenes as a target system for quantum simulation. We demonstrated simulation of the canonical 20-spin dodecahedron and 60-spin buckyball Ising cages. Classically, we performed MPS time-evolution simulations for both, increasing bond dimension to 724, until the simulations no longer fit in memory for a single GPU in the Frontier supercomputer. (While geometry-specific tensor network methods may be effective in the short-ta\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$t_{a}$\end{document} regime [37, 38], we found that the belief-propagation-based time-evolution method of Tindall et al. [39] struggles on the simulations performed here (see Fig. S4 of SM).) Such fullerene structures can scale arbitrarily large, with numerous isomers of equivalent size [24], enabling a wide variety of simulations even for purely AFM systems of modest size. Simulations of such models have been shown to be achievable on existing neutral-atom platforms [21]. Precise reproduction of the inherent symmetries and degeneracies of these systems is crucial for achieving high-quality simulation results, as they enable the emergence of correlations from the extensive classical ground-state manifold when quantum fluctuations are introduced. Additionally, these symmetries play a vital role in mitigating sampling errors that can arise during Monte Carlo inference in quantum simulators, particularly for platforms such as trapped-ion or neutral-atom systems, where shot rates are significantly slower than those of superconducting platforms. Our results delineate the boundary between coherent quantum evolution and thermally dominated dynamics in current quantum annealers. At short to intermediate annealing times, the system evolves approximately unitarily and exhibits quantum fluctuation–induced selection, as evidenced by the nontrivial suppression of floppy dimers in agreement with second-order perturbative predictions. As the annealing time increases, however, the system transitions into a thermal regime where decoherence dominates, and the output distribution shifts toward classical low-energy configurations, deviating from the structure of the quantum ground state . This transition is clearly observed in both residual energy and structural observables. While classical methods such as MPS can accurately reproduce the target dynamics for modest system sizes, they become computationally intractable for larger, more connected geometries. In this regime, frustrated fullerene spin systems provide a controlled, symmetry-rich platform for assessing the onset of thermalization and for evaluating the fidelity of quantum devices in capturing fine-grained many-body quantum effects. In our experiments, we observed significant performance improvements across successive generations of superconducting quantum annealers, demonstrating the potential for using these systems to probe quantum phenomena and enhance emerging technologies. Ultimately, this work takes a modest but significant step toward a future scenario where classical methods are inadequate, and quantum simulators must rely on cross-platform testbeds for mutual validation. Supplementary Information Below is the link to the electronic supplementary material. (PDF 465 kB) (PDF 465 kB)