Experimental establishment of nonlinear cascaded grid-computing layers

Establish experimentally whether multiple cascaded physical-computing layers with trainable nonlinear transformations can extend the linear steady-state electric-grid operator, potentially by exploiting the switching and saturation characteristics of power-electronic converters.

Background

The demonstrated computational substrate is a steady-state electric-grid operator whose input-output transformation is linear. The discussion identifies cascaded physical layers with trainable nonlinear transformations as a distinct extension that could increase the expressivity of infrastructure-native computing.

Potential nonlinear mechanisms include switching and saturation in power-electronic converters. The paper does not establish experimentally whether such mechanisms can be integrated into cascaded grid-computing architectures or trained effectively.

References

Second, the steady-state physical operator considered here is linear. Extending the architecture to multiple cascaded physical layers with trainable nonlinear transformations would require nonlinear physical mechanisms, potentially including the switching and saturation characteristics of power-electronic converters. Such an extension is conceptually distinct from the present demonstration and remains to be established experimentally.

— Infrastructure-Native Computing with Electric Power Grids  (2610.08390 - Song et al., 6 Oct 2026) in Discussions, paragraph beginning “Several limitations define the next stage of the work”