In a new study, Pushing the Classical Frontier of 1D Fermi–Hubbard Quench Dynamics Beyond Current Quantum Simulations, the team developed a highly optimized simulation framework combining U(1)×SU(2) symmetry-preserving tensor networks, GPU acceleration, and advanced Time-Dependent Variational Principle (TDVP) algorithms to tackle one of the most demanding benchmarks in quantum simulation.
Using up to four NVIDIA H200 GPUs, the researchers reached bond dimensions of approximately 62,000, among the largest ever reported for TDVP simulations of real-time quantum dynamics. This enabled fully converged simulations of the one-dimensional Fermi–Hubbard model in a regime that had previously remained unresolved by classical approaches.
The work provides the first rigorous classical certification of the high-entanglement regime of a recent large-scale quantum hardware experiment involving a 120-qubit simulation of Fermi–Hubbard dynamics. Beyond reproducing and validating the complete experimental time window, the Multiverse team extended the simulations to t = 7, beyond the range reached by the quantum processor itself.
The results demonstrate how advances in algorithms, symmetry exploitation, and modern GPU architectures continue to expand the limits of classical simulation. They also highlight the importance of continuously reassessing quantum advantage claims against the strongest available classical methods.
This achievement further strengthens Multiverse Computing’s leadership in large-scale tensor-network algorithms, quantum simulation, and high-performance scientific computing, while contributing important new benchmarks for the quantum computing community.
Read the full paper here: https://arxiv.org/abs/2606.04771