I have googled the relationship between tree depth, leapfrog, acceptance rate, and running time. But I did not find a solution any way, so can anyone help me with my problem below:

The same model with different number of items was estimated by a hierarchical model. The same specification was used with max_treedepth = 10, adapt_delta=0.90. One instance ran so much faster, with the following running time:

Chain 1: 1000 transitions using 10 leapfrog steps per transition would take 630 seconds.

Chain 1: Adjust your expectations accordingly!

While another instance ran so slowly, which takes about 20k seconds for 1000 transitions.

What can I do to make the second instance run faster? How do I adjust leapfrog, beside treedepth, in rstan?

Thank you very much.