Hamiltonian Monte Carlo: why gradients beat a random walk in 100 dimensions

Afterwards you can explain how Hamiltonian Monte Carlo uses gradients and leapfrog steps to move far, tune its step size and path length, and spot a divergence.