arXiv:2609.26516
Tensor Network Simulation of Dynamic Circuits
Innocenzo Fulginiti, Alessandro Poggiali, Christian B. Mendl
Abstract
Dynamic circuits, which incorporate mid-circuit measurements and classically controlled operations, extend the expressive power of quantum programs and are central to applications such as quantum error correction, state preparation, and distributed quantum algorithms. However, their classical simulation is challenging due to the exponential growth of execution branches induced by mid-circuit measurements. In this work, we develop a tensor network approach for simulating dynamic circuits by extending a DMRG-based method for quantum circuit simulation to support dynamic circuit operations. To address the exponential proliferation of execution branches, we introduce and compare two strategies: a single-path stochastic approach that samples individual measurement outcomes, and a multi-path approach that maintains an ensemble of branches. We benchmark these methods on standard dynamic circuits, including the teleportation protocol and GHZ state preparation, as well as on random dynamic circuits. Our analysis explores the trade-offs between simulation accuracy and computational efficiency. These results provide a practical and scalable framework for the classical simulation of dynamic circuits.
AI Summary
The paper addresses classical simulation of dynamic quantum circuits with mid-circuit measurements, resets, and classical feedforward. It extends an existing DMRG-inspired variational circuit simulator using matrix product states (MPS) and tree tensor networks (TTN). Unitary gates are grouped into chunks and variationally compressed; chunks terminate at measurements, whose probabilities and conditional states are obtained by tensor-network contraction and local projection. Two workflows handle measurement histories: single-path simulation samples one outcome at each measurement, while multi-path simulation explicitly propagates branches and optionally retains only the $M$ highest-probability histories. Without pruning, $r$ measurements can produce $2^r$ branches. With a branch cap, runtime and memory scale linearly in $M$, in addition to the underlying tensor-network costs. Single-path simulation stores one state but requires repeated runs for ensemble statistics. Validation uses teleportation, dynamic GHZ preparation on 5–17 qubits, and 27-qubit random dynamic circuits. Teleportation reproduces the expected branch probabilities; GHZ tests show how branch truncation and sampling affect total variation distance. Random-circuit tests use bond dimensions $\chi\in\{8,16,32\}$, depths 2–8, ten single-path runs, and $M=8$. Larger bond dimensions improve the reported fidelity estimates, while TTN is faster and MPS uses less memory in these tests. Fidelity is estimated from products of compression overlaps rather than exact final-state comparisons. The stated approximation sources are tensor-network compression and branch pruning. Further work includes larger and deeper benchmarks, adaptive pruning, other network architectures, and understanding the interaction between entanglement and network geometry.
Central claim may not hold: The claimed scalable accuracy is conditional on manageable bond dimensions and retained probability mass, with no overall error guarantee or accuracy-preserving scaling demonstration. The random-circuit fidelity results use an unvalidated product of compression overlaps and therefore do not establish final-state or full-ensemble simulation accuracy. The teleportation benchmark's computational-basis measurements cannot verify preservation of the teleported state's relative phase. The described GHZ circuit resets (n−1)/2 of its n qubits, precluding an n-qubit GHZ output as described, and its path-distribution tests do not verify GHZ coherence. The displayed local variational update requires an isometric canonical environment that is not specified, so its derivation is incomplete for general MPS and TTN gauges.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- Accepted at CASCON 2026 conference