machine learning - I found the optimal number of nodes in a single hidden layer. Should I keep this number constant when adding a second hidden layer? -

i've done grid search in order find optimal number of nodes in deep network using auroc measure optimality. let's having 100 nodes in first hidden layer produces highest auroc value of 0.7. can assume when adding second hidden layer having 100 nodes in first hidden layer lead best model? don't want grid search second hidden layer while varying number of nodes in first , second hidden layers because add more hidden layers, lead exponentially longer run times.

"of course not!. there tell wrong assumption, recommend read basic introduction neural networks (given nonlinearities). i'm pretty sure, our world different if assumption true (nn-architecture-optimization in polynomial-time; implication of assumption me)"- courtesy of sascha. don't mean plagiarize answer, want others see correct answer!