Fig. 2.

Schematic architecture of DeepONet. DeepONet learns the mapping operator $G$ from an input function $g$ to its corresponding output function $G\left(g\right)$. The input of branch and trunk net are $g$ and $y\in {ℝ}^p$, which represents a discretized function from $g\left(x_1\right)$ to $g\left(x_m\right)$ and information such as coordinates and time, respectively. Note that the output dimension of the trunk net, $n$, is consistent with that of the branch net. The final output $G\left(g\right)\left(y\right)$ is computed as the dot product of $b$ and $t$