Computational Science Asked on October 23, 2021
I have a question about the Markov Chain Hamiltonian Monte Carlo (MCHMC).
Hamiltonian Monte Carlo is known as Hybrid Monte Carlo too. I’ll describe the steps of the algorithm.
We have at the beginning a coordinates configuration {$q_i$} and generate momenta configuration {$p_i$} from a gaussian distribution at the desired temperature $T$.
$i$ run over the degrees of freedom of the system. So we initially have the configuration {$q_i, p_i$}. This is the first configuration in our Markov chain
We make $n$ leapfrog steps and obtain the configuration {$q’_i, p’_i$}
We accept or reject this last configuration using a Metropolis test
Now we continue with an example. Suppose this last configuration is rejected. Then we keep the old configuration {$q_i,p_i$} (first question: this is the second configuration of our Markov chain, right?) and generate (for the second time) a new {$p”_i$} set of momenta gaussian distributed. So we now have the configuration {$q_i, p”_i$}.
Now we make other $n$ leapfrog steps and arrive at the configuration {$q”_i, p”’_i$}. Second question: if Metropolis rejects it, the third configuration of our Markov chain will be {$q_i, p_i$} or {$q_i, p”_i$}?.
Now, instead, we suppose Metropolis accepts {$q”_i, p”’_i$}, and we generate a new gaussian set {$p^{IV}_i$}. So we now have the configuration {$q”, p^{IV}_i$} and then continue from it with leapfrog. The last question is: which is the third configuration of our Markov chain, {$q”_i, p”’_i$} or {$q”_i, p^{IV}_i$}?
The algorithm for perfoming a single HMC step is as follows:
Input: Some initial configuration $vec{y}_i$ and momentum $vec{p}_i$.
Output: Next configuration $vec{y}_{i+1}$ and momentum $vec{p}_{i+1}$
Now, to answer your questions (and to give some additional remarks):
Answered by cos_theta on October 23, 2021
Get help from others!
Recent Answers
Recent Questions
© 2024 TransWikia.com. All rights reserved. Sites we Love: PCI Database, UKBizDB, Menu Kuliner, Sharing RPP