A posteriori certification for physics-informed neural networks

We propose rigorous lower and upper bounds for a PINN approximation to PDEs by efficiently computing the Riesz representations of suitable extension and restrictions of the PINN residual towards geometrically simpler domains, which are either embedded or enveloping the original domain. Error bounds are proven and detailed for elliptic as well as parabolic problems. Numerical experiments show the good quantitative behaviour of the derived upper and lower error bounds.