General proof of local observability

Prove local observability for general coupled physical systems and sensor configurations, beyond the specific configurations for which the physics-constrained attentive neural operator framework provides numerical verification.

Background

The paper derives conditional stability and parameter-recovery estimates under local state-stability and sensor-observability assumptions. For the numerical benchmarks, local parameter-to-observation sensitivity is checked computationally, including a finite-difference diagnostic for the two-dimensional lid-driven cavity problem at Reynolds number 5000.

The authors do not establish observability theoretically for arbitrary coupled systems, sensor layouts, or parameter regimes. Consequently, extending the numerical evidence to a general proof of local observability remains unresolved and is important for determining when sparse single-field observations uniquely constrain unobserved fields and unknown physical parameters.

References

We verify local observability numerically for the tested configurations rather than proving it for general systems, and we optimize each case separately, requiring retraining for every new configuration.