1: \begin{abstract}

2: We construct and analyze approximation rates of

3: deep operator networks (ONets) between infinite-di\-men\-sional spaces

4: that emulate with an exponential rate of convergence the coefficient-to-solution

5: map of elliptic second-order partial differential equations. 

6: In particular, 

7: we consider problems set in $d$-dimensional periodic domains, $d=1, 2, \dots$, 

8: and with analytic right-hand sides and coefficients. 

9: Our analysis covers linear, elliptic second order divergence-form PDEs as, e.g.,

10: diffusion-reaction problems, parametric diffusion equations, and elliptic systems 

11: such as linear isotropic elastostatics in heterogeneous materials.

12: 

13: We leverage the exponential convergence of spectral collocation methods for

14: boundary value problems whose solutions are analytic. In the present periodic

15: and analytic setting, this follows from classical elliptic regularity. Within

16: the ONet branch and trunk construction of Chen and Chen \cite{ChenChen1993} and

17: of Lu et al.~\cite{lu2020deeponet}, we show the existence of deep ONets which

18: emulate the coefficient-to-solution map to a desired accuracy in the $H^1$

19: norm, uniformly over the coefficient set. 

20: We prove that the neural networks in

21: the ONet have size $\mathcal{O}(\left|\log(\varepsilon)\right|^\kappa)$,

22: where $\varepsilon>0$ is the approximation accuracy, 

23: for some $\kappa>0$ depending on the physical space dimension.

24: \end{abstract}

25: