1: \begin{abstract}

2: Hamiltonian Monte Carlo (HMC) is an efficient method of simulating smooth

3:  distributions and has motivated the widely used No-U-turn Sampler (NUTS) and

4:  software Stan.  We build on NUTS and the technique of ``unbiased sampling''

5:  to design HMC algorithms that produce perfect simulation of general

6:  continuous distributions that are amenable to HMC.  Our methods enable

7:  separation of Markov chain Monte Carlo convergence error from experimental

8:  error, and thereby provide much more powerful MCMC convergence diagnostics

9:  than current state-of-the-art summary statistics which confound these two

10:  errors.  Objective comparison of different MCMC algorithms is provided by the

11:  number of derivative evaluations per perfect sample point.  We demonstrate

12:  the methodology with applications to normal, $t$ and normal mixture

13:  distributions up to 100 dimensions, and a 12-dimensional Bayesian Lasso

14:  regression.  HMC runs effectively with a goal of 20 to 30 points per

15:  trajectory.  Numbers of derivative evaluations per perfect sample point range

16:  from 390 for a univariate normal distribution to 12,000 for a 100-dimensional

17:  mixture of two normal distributions with modes separated by six standard

18:  deviations, and 22,000 for a 100-dimensional $t$-distribution with four

19:  degrees of freedom.

20: % Exact numbers 388, 12061, 21673

21: \end{abstract}

22: