The Signal
Preprint reports Pauli simulator scaling beyond 10¹¹ strings
A November 12 preprint presents an open-source Pauli-propagation implementation and reports simulations with more than 10¹¹ Pauli strings. Its partitioning method limits which groups of strings need to exchange data during gate application.

An open-source implementation reported in a November 12 preprint can simulate quantum dynamics with more than 1011 Pauli strings, according to its authors. Their method partitions strings using a binary linear hash, limiting the data exchanges needed when applying gates across processors and compute nodes.
Source arxiv.org
Partitioning to contain communication
The preprint, by Alev Orfi, Andrew Kille, J. Clayton Peacock, Jack Dunham and Lukas Devos, is a first-version paper on arXiv. It describes paulistrings-rs, which the authors benchmarked on a kicked Ising circuit and quenches of the two-dimensional transverse-field Ising model.
Pauli propagation represents an observable as a sum of Pauli strings and evolves that sum through a circuit or dynamical process. The paper says the operations on individual strings are relatively cheap; memory bandwidth and capacity, along with the need to combine duplicate strings after gates, constrain larger simulations.
In the proposed approach, a linear hash assigns strings to partitions. For a gate or channel acting on k qubits, the authors show that each partition can couple to at most 4k partitions, fixed by the gate. Those partitions form independent tasks that can be updated and deduplicated without communication with other task groups. For Pauli rotations, the paper says each partition interacts with at most one other partition.
Benchmarks across nodes
For the kicked Ising test, the implementation propagated an observable through a 127-qubit circuit and calculated a single-site magnetization after 20 Trotter steps. On one two-socket node, scaling was close to ideal up to 16 threads; at 96 threads, reported parallel efficiency was 61% for one truncation threshold and 38% for another. The paper says the decline was consistent with available memory bandwidth becoming saturated.
In a separate weak-scaling test on the two-dimensional model, the authors held the Pauli-string budget per node fixed as they increased the node count. At 32 nodes, they report efficiencies between 68% and 74%, depending on the per-node budget. The largest configuration supported up to 1.6 × 1011 strings.
Large expansions and limits
For a further test, the team examined a 14-by-14 transverse-field Ising quench using coefficient thresholds from 10−4 to 2.5 × 10−7. At the smallest threshold, the expansion reached 1.64 × 1011 strings across 64 nodes. At tJ = 1, the authors report that this calculation retained 99.1% of the exact squared operator norm.
The comparison with matrix-product and tensor-network calculations agreed within those methods’ stated convergence limits. At later times, the results diverged; the authors say the reference calculations no longer provide a converged benchmark for quantitative accuracy. They identify extending the approach to GPU-based implementations as a possible next step, while noting that distributed runs still require pairwise exchanges and synchronization for gates that couple workers.
Sources
Related stories
- Researchers Pre-Train 64-Qubit Circuits with Pauli Propagation
- Preprint reports Rigetti processor simulation of inner-product change
- 63-qubit experiment tracks electric strings in gauge theory
- Nature study reports resistance minimum in superfluid channels
- Physical Review Applied publishes classical wavebit circuit emulator