Floquet Ising dynamics on 51- and 74-qubit heavy-hex lattices

A periodically kicked Ising magnet on IBM's heavy-hex lattice shows long-lived oscillations over tens of drive cycles before heating sets in. Error-mitigated IBM hardware has measured them with percent-level precision, but tensor-network and Pauli-path simulations of the same circuits have not converged, so a converged classical answer would confirm or overturn the hardware data.

Task

Start $N$ qubits on a heavy-hex patch in $|0\rangle^{\otimes N}$ and apply $N_c$ Floquet cycles $U_F=U_3U_2U_1$ , where

U_m=\prod_{(j,k)\in\mathcal E_m}R_{ZZ}(\theta_{zz})\;\prod_i R_Z(\theta_z/3)\;\prod_i R_X(\theta_x/3),

the sets $\mathcal E_1,\mathcal E_2,\mathcal E_3$ partition the edges of the patch, $\theta_x=1.56567$ , $\theta_z=0.33879$ and $\theta_{zz}=\pi/3$ . Estimate classically, after the last cycle, the magnetization $M=\frac1N\sum_i\langle Z_i\rangle$ or the correlator $ZZ_d$ , the average of $\langle Z_iZ_j\rangle$ over pairs at graph distance $d$ .

Instance ids are floquet_mixed_field_ising_ plus observable, $N$ and $N_c$ . All five are active candidates on the tracker: mag_51qx16c, zzd2_51qx16c and zzd3_51qx16c ( $M$ , $ZZ_2$ , $ZZ_3$ at 16 cycles), mag_51qx30c and mag_74qx27c. The QASM files, edge layers and pair lists are in the tracker folder.

Background

Driven many-body systems tend to absorb energy until local observables look like infinite temperature, but they can first spend a long time in a prethermal state [1]. Qedma, IBM, RIKEN and BlueQubit posted these circuits in May and July 2026. In the accompanying paper, Leviatan et al. (2026) use Qedma's QESEM error mitigation [2] on the ibm_boston processor and report long-lived subharmonic oscillations in systems of up to 74 qubits [1]. Kicked Ising circuits on heavy-hex lattices are familiar from IBM's 2023 utility experiment [4], and the ORQA sparse Pauli-path code used here was benchmarked on 127-qubit kicked Ising circuits like that experiment's [3].

Data and evaluation

The tracker records a value with optional lower and upper bounds, runtime and hardware, and compares results only within one instance.
QESEM (unbiased) uses quasi-probabilistic mitigation, and its error bars are the statistical uncertainty of that estimator [2]; later cycles use a zero-noise-extrapolation variant, QESEM (ZNE).
The tracker does not rank submissions by a score. The metric for this entry is our choice, not the tracker's: the width of a defensible error bar on the expectation value, smaller is better, backed by evidence of convergence. There is no baseline score: the tracker shows hardware intervals but no agreed classical value.

What is known

Accepted tracker entries exist so far only for zzd3_51qx16c: QESEM (unbiased) 0.2111 in [0.1984, 0.2238], and sparse Pauli paths (ORQA, threshold $5\times10^{-8}$ , weight cap 16) 0.1834, both from Qedma, IBM, RIKEN and BlueQubit (2026). The tracker's instance README gives the magnetization:

51 qubits, cycle 16: QESEM (unbiased) 0.3904 [0.3691, 0.4117]; PEPS with belief propagation (PEPS-BP, bond dimension 700) 0.4418; sparse Pauli paths 0.4111 and 0.3870 for two truncations.
51 qubits, cycle 30: QESEM (ZNE) 0.3662 [0.3596, 0.3727]; PEPS-BP 0.5411; sparse Pauli paths 0.3851 and 0.3498.
74 qubits, cycle 27: QESEM (ZNE) 0.3653 [0.3621, 0.3685]; no classical estimate.

According to that README, PEPS-BP is reliable only to about 12 cycles, and the Pauli-path runs, which tracked up to about $1.4\times10^{12}$ paths on the Fugaku supercomputer, disagree with each other and with the hardware beyond 7 cycles. In the tracker review, a reviewer argued that larger bond dimensions or a tensor-network operator evolution might converge at 16 cycles, and that the true value can lie outside the hardware interval; the submitters replied that PEPS-BP is unconverged there and about 2.5 standard deviations from the unbiased QESEM value. Submissions for the other four instances are still under review.

What would count as progress

A classical magnetization for 51 qubits at 16 or 30 cycles, converged in bond dimension or truncation, with an error bar comparable to QESEM's (interval widths about 0.04 and 0.013), landing inside the hardware interval or convincingly outside it.
A first classical estimate for the 74-qubit, 27-cycle instance.
Converged $ZZ_2$ and $ZZ_3$ correlators at 16 cycles.

A result can be checked by rerunning the submitted code on the published QASM files.

References

E. Leviatan et al., "Resolving Structure in Prethermal Floquet Dynamics with Precision Quantum Computation", preprint (2026). arXiv:2607.24937
D. Aharonov et al., "Reliable high-accuracy error mitigation for utility-scale quantum circuits", preprint (2025). arXiv:2508.10997
L. Broers, R.-Y. Sun and S. Yunoki, "Scalable simulation of quantum many-body dynamics with or-represented quantum algebra", Phys. Rev. Applied 26, 024046 (2026). doi:10.1103/y8ft-m61w
Y. Kim et al., "Evidence for the utility of quantum computing before fault tolerance", Nature 618, 500–505 (2023). doi:10.1038/s41586-023-06096-3

Source

Imported from Quantum Advantage Tracker floquet_mixed_field_ising, curated by Quantum Advantage Tracker reviewers (IBM, Flatiron Institute, BlueQubit, Algorithmiq and others). Open according to the source as of 2026-10-20. The statement above is written in our own words; see the source for the original wording, history and discussion.