Problem constructing a suitable NN architecture to solve partial differential equations with non-separable solutions

[AneleNcube]

Have physics-informed neural networks been shown to solve partial differential equations whose solutions are non-separable functions? What suitable setups work?: deep/shallow NNs, what type of activation functions?, etc.

[vl-dud]

Check examples, there are non-separable solutions