Abstract
We propose PO-CKAN, a physics-informed deep operator framework based onChunkwise Rational Kolmogorov--Arnold Networks (KANs), for approximating thesolution operators of partial differential equations. This framework leveragesa Deep Operator Network (DeepONet) architecture that incorporates ChunkwiseRational Kolmogorov-Arnold Network (CKAN) sub-networks for enhanced functionapproximation. The principles of Physics-Informed Neural Networks (PINNs) areintegrated into the operator learning framework to enforce physicalconsistency. This design enables the efficient learning of physicallyconsistent spatio-temporal solution operators and allows for rapid predictionfor parametric time-dependent PDEs with varying inputs (e.g., parameters,initial/boundary conditions) after training. Validated on challenging benchmarkproblems, PO-CKAN demonstrates accurate operator learning with results closelymatching high-fidelity solutions. PO-CKAN adopts a DeepONet-style branch--trunkarchitecture with its sub-networks instantiated as rational KAN modules, andenforces physical consistency via a PDE residual (PINN-style) loss. On Burgers'equation with $\nu=0.01$, PO-CKAN reduces the mean relative $L^2$ error byapproximately 48\% compared to PI-DeepONet, and achieves competitive accuracyon the Eikonal and diffusion--reaction benchmarks.