Hybrid Monte Carlo on Hilbert spaces (Q719371)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Hybrid Monte Carlo on Hilbert spaces
scientific article

    Statements

    Hybrid Monte Carlo on Hilbert spaces (English)
    0 references
    10 October 2011
    0 references
    The authors consider the problem of sampling a probability measure \(\pi\) on a Hilbert space defined via the density with respect to a Gaussian measure \(\pi_{0}\): \[ \frac{d\pi}{d\pi_{0}}(q)\propto\exp(-\Phi(q)). \] Any algorithm designed to sample \(\pi\) should be implemented on a finite-dimensional space of dimension \(N\). The number of steps required to explore the target distribution \(\pi\) typically grows with \(N\). The authors propose a generalized hybrid Monte Carlo algorithm which overcomes these shortcomings. They develop the following issues in the infinite-dimensional setting: (i) construction of a probability measure \(\Pi\) in an enlarged phase space having the target \(\pi\) as a marginal together with a Hamiltonian flow that preserves \(\Pi\); (ii) development of a geometric numerical integrator for the Hamiltonian flow; (iii) derivation of an accept/reject rule to ensure preservation of \(\Pi\) when using the above integrator instead of the actual Hamiltonian flow. The standard HMC algorithm was introduced in [\textit{S. Duane, A. D. Kennedy, P. Pendleton} and \textit{D. Roweth}, ``Hybrid Monte Carlo'', Phys. Lett. B 195, No.~2, 216--222 (1987; \url{doi:10.1016/0370-2693(87)91197-X})].
    0 references
    0 references
    0 references
    0 references
    0 references
    Hamiltonian dynamics
    0 references
    splitting technique
    0 references
    absolute continuity
    0 references
    hybrid Monte Carlo
    0 references
    0 references
    0 references
    0 references
    0 references