Diagnosing Overfitting and Choosing Regularization Parameters

Your LightGBM model shows this training curve:

Round  | Train RMSE | Val RMSE
-------|------------|----------
100    |  0.45      |  0.47
500    |  0.31      |  0.36
1000   |  0.22      |  0.38
1500   |  0.15      |  0.42
2000   |  0.09      |  0.48

1. What is happening and at approximately which round is the optimal model?
2. Which parameters are responsible for allowing the model to overfit this way?
3. Rank the regularization parameters by their expected impact and explain what each one does to the model's structure.