Using the Euclidean-Gaussian kinetic energy, you can set the inverse Euclidean metric to the target covariances (estimated during tuning, e.g. tune=1000). The kinetic energy is computed as 0.5 * p' * M^-1 * p + log|M| + const, if you only take the diagonal of M^-1 that is v above. The momentum is given as Normal(p | 0, M), which you scale with the inverse of the potential vector as above. You can find more information in Betancourt’s paper (chapter 4.2).