1: \begin{abstract}

2: We study an infinite-dimensional optimization problem that aims 

3: to identify the Nemytskii operator in the nonlinear part of a

4: prototypical semilinear elliptic partial differential equation (PDE) which 

5: minimizes the distance between the PDE-solution and a given desired state.

6: In contrast to previous works, we 

7: consider this identification problem in a low-regularity regime in which the 

8: function inducing the Nemytskii operator is a-priori only known to be 

9: an element of $H^1_{loc}(\R)$. This makes the studied problem class a

10: suitable point of departure for the rigorous analysis of training problems for learning-informed PDEs 

11: in which an unknown superposition operator is approximated 

12: by means of a neural network with nonsmooth activation functions (ReLU, leaky-ReLU, etc.).

13: We establish that, despite the low regularity of the controls, 

14: it is possible to derive a classical 

15: stationarity system for local minimizers and to solve the considered problem by means of 

16: a gradient projection method.

17: The convergence of the resulting algorithm is proven in the function space setting. 

18: It is also shown that the established first-order necessary optimality conditions imply that 

19: locally optimal superposition operators share various characteristic properties 

20: with commonly used activation functions: They are always sigmoidal, 

21: continuously  differentiable away from the origin, and typically possess a distinct kink 

22: at zero. The paper concludes with numerical experiments which 

23: confirm the theoretical findings.

24:  

25: \end{abstract}