arXiv:2610.03254
Quantum approaches for particle-in-cell codes
Ryan J. J. Connor, Andrew J. Daley, Callum W. Duncan, Preetma Soin
Abstract
Solving the high-dimensional, nonlinear partial differential equations that govern the rich physics of plasmas remains a major computational challenge. In fluid dynamics, quantum-inspired methods based on tensor-network techniques have recently emerged as a promising new paradigm, offering large degrees of data compression on conventional hardware and a natural pathway towards solving exponentially large problems on quantum computers. Early studies suggest that tensor-network methods may also help overcome key bottlenecks in computational plasma dynamics, particularly for industry-relevant scenarios where conventional simulations remain costly. In this work, we present recent progress in developing tensor-network methods compared to standard particle-in-cell codes and discuss their capability to overcome the bottlenecks of conventional approaches. We further extend the tensor-network formulation to support the insertion of electromagnetic fields at arbitrary spatial locations. This capability is demonstrated through the injection of a laser field to accelerate a plasma, providing a key step towards tensor-network-based simulation of externally driven plasma systems.
AI Summary
The paper addresses how to inject electromagnetic fields at interior locations in tensor-network simulations of plasma dynamics, extending earlier implementations restricted to boundary injection. It uses a classical matrix-product-state (MPS) representation of discretized Vlasov–Maxwell fields and distribution functions, with binary grid encoding and finite-difference derivatives represented as matrix-product operators. The new component is a total-field/scattered-field (TFSF) construction. The domain is partitioned into regions containing either the total field or only scattered fields. At their interface, analytical incident-field contributions correct the finite-difference derivatives. These corrections are implemented using MPS tensors supported at the two adjacent grid points. The construction also accommodates two opposing injected beams. A Maxwell-field propagation example places injection interfaces 32 grid points from the boundaries and shows field snapshots at $t=L_x/(6c)$ and $t=L_x/c$. A beat-wave calculation measures average runtime per step over the first 50 steps at fixed bond dimension $\chi=512$; the reported scaling indicates that adding TFSF preserves logarithmic runtime dependence on grid resolution in this test. Earlier bump-on-tail and beat-wave results are reviewed as background, including the need for $\chi=512$ to reproduce beat-wave current density without noticeable truncation errors. The stated efficiency depends on restricting correlations without losing relevant physics. Further development is needed to match mature particle-in-cell capabilities. The paper does not present a quantum-computer implementation; its quantum connection is through tensor-network representations and possible future algorithms.
Central claim may not hold: The abstract's claimed demonstration of laser-driven plasma acceleration is unsupported by the reported new results, which show electromagnetic-field propagation and runtime measurements rather than plasma acceleration. The fixed-bond-dimension benchmark over the first 50 time steps does not establish logarithmic simulation cost at a specified accuracy as grid resolution or simulation duration increases. The Outlook reverses the grid-size relationship: the number of MPS tensors is logarithmic in the represented grid size, not vice versa.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 4 pages, 7 figures. Accepted Workshop Paper (QC4PDE) at IEEE Quantum Week 2026