Speaker
Description
Superconducting circuits containing Josephson junctions are important for both fundamental research and applications ranging from precision metrology to quantum technologies. Even relatively simple circuits exhibit rich nonlinear dynamics, including phase slips and switching between different dynamical regimes [1]. Such systems are often well described by the resistively and capacitively shunted junction (RCSJ) model and its extensions. However, reliable models of more complex and noisy circuits are often difficult to formulate. In this work, we employ neural stochastic differential equations [2], a machine-learning approach that combines known physical models with neural networks to describe such systems. The neural network learns the missing contributions to the dynamics, including stochastic effects, while preserving consistency with the underlying physics. The approach is sufficiently flexible to describe different circuit architectures and may provide a general framework for modeling complex superconducting circuits. As a proof of concept, we apply the method to RCSJ models with different Josephson potentials [3,4] and a more complex circuit.
References:
[1] M. Žonda, W. Belzig, T. Novotný, Phys. Rev. B 91 (13), 134305 (2015)
[2] X. Li, T.-K. L. Wong, R. T. Q. Chen, D. Duvenaud, Proc. Mach. Learn. Res. 108, 3870-3882 (2020)
[3] M. Žonda, W. Belzig, E. Goldobin, T. Novotný, Phys. Rev. B 110 (5), 054306 (2024)
[4] F. Dominguez, F. Hassler, G. Platero, Phys. Rev. B 86, 140503(R) (2012)