Losses start very high and drop to near zero extremely quickly

Hi all,

I’m training a network to solve a system of partial differential equations, and the loss is extremely large at iteration 1000, by iteration 2000 it’s already down to almost zero and it stays down, although it decreases incrementally. Clearly there is something wrong, but I can’t narrow down what it is. I imagine that something like this is probably symptomatic of some kind of pathology that many of you would have seen/know about, that it’s a tell-tale sign of something specific that I’m messing up. I’d very grateful for any suggestions on what you’d look at.

Many thanks,