Symmetry-driven in-situ physical adjoint computing beyond reciprocity
Abstract
Nonreciprocity enriches wave control and information processing through directional transport, yet challenges conventional in-situ physical adjoint optimization, which relies on reciprocity to acquire adjoint responses. Here we demonstrate symmetry-driven in-situ physical adjoint computing (Sym-iPAC) in nonreciprocal systems, which enables access to adjoint responses on the original hardware through three classes of transpose symmetry: local geometric permutations, complex gauge reweighting, and latent symmetry in effective port dynamics revealed by Schur reduction. Experiments with programmable nonreciprocal microwave networks validate these transpose relations and yield adjoint gradients, enabling closed-loop optimization. These results establish experimentally accessible transpose symmetry as a route to in-situ physical adjoint computing beyond reciprocity.
Related
- Reciprocity Separates Gradient Flow from Rotation in Conservative Physical Learning
- Physical policy gradient theorem for in situ stochastic-adjoint training
- Physical Adjoint Learning in Wave Systems with Asymmetric Couplings
- Aspects of Symmetry in Nuclear Quantum Computing Variational Algorithms
- Heterogeneous architectures enable a 138x reduction in physical qubit requirements for fault-tolerant quantum computing under detailed accounting
BibTeX
@misc{li2026symmetrydriven,
title = {{Symmetry-driven in-situ physical adjoint computing beyond reciprocity}},
author = {Cunshen Li and Rong Li and Jiongning Che and Cheng-Zhen Wang and Liang Huang},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.34637},
eprint = {2610.34637},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.34637},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Symmetry-driven in-situ physical adjoint computing beyond reciprocity
AU - Cunshen Li
AU - Rong Li
AU - Jiongning Che
AU - Cheng-Zhen Wang
AU - Liang Huang
PY - 2026
DA - 2026/10/25
PB - arXiv
DO - 10.48550/arxiv.2610.34637
UR - https://arxiv.org/abs/2610.34637
AB - Nonreciprocity enriches wave control and information processing through directional transport, yet challenges conventional in-situ physical adjoint optimization, which relies on reciprocity to acquire adjoint responses. Here we demonstrate symmetry-driven in-situ physical adjoint computing (Sym-iPAC) in nonreciprocal systems, which enables access to adjoint responses on the original hardware through three classes of transpose symmetry: local geometric permutations, complex gauge reweighting, and latent symmetry in effective port dynamics revealed by Schur reduction. Experiments with programmable nonreciprocal microwave networks validate these transpose relations and yield adjoint gradients, enabling closed-loop optimization. These results establish experimentally accessible transpose symmetry as a route to in-situ physical adjoint computing beyond reciprocity.
ER -