“…[20] recently proposed to use a stochastic differential equation (SDE) instead of an ODE. For the unconstrained case the use of SDEs was also advocated in [21,22,23,24]. The advantage of the SDE is that it can help the algorithm escape from local minima and eventually reach the global solution.…”
Section: Diffusions For Constrained Global Optimizationmentioning
confidence: 99%
“…This function is usually referred to as the annealing schedule. In order for the algorithm to theoretically exhibit convergence to the global solution, the annealing schedule is selected as follows [21,23,22,14]:…”
Section: Diffusions For Constrained Global Optimizationmentioning
We propose a stochastic algorithm for the global optimization of chance constrained problems. We assume that the probability measure with which the constraints are evaluated is known only through its moments. The algorithm proceeds in two phases. In the first phase the probability distribution is (coarsely) discretized and solved to global optimality using a stochastic algorithm. We only assume that the stochastic algorithm exhibits a weak* convergence to a probability measure assigning all its mass to the discretized problem. A diffusion process is derived that has this convergence property. In the second phase, the discretization is improved by solving another nonlinear programming problem. It is shown that the algorithm converges to the solution of the original problem. We discuss the numerical performance of the algorithm and its application to process design.
“…[20] recently proposed to use a stochastic differential equation (SDE) instead of an ODE. For the unconstrained case the use of SDEs was also advocated in [21,22,23,24]. The advantage of the SDE is that it can help the algorithm escape from local minima and eventually reach the global solution.…”
Section: Diffusions For Constrained Global Optimizationmentioning
confidence: 99%
“…This function is usually referred to as the annealing schedule. In order for the algorithm to theoretically exhibit convergence to the global solution, the annealing schedule is selected as follows [21,23,22,14]:…”
Section: Diffusions For Constrained Global Optimizationmentioning
We propose a stochastic algorithm for the global optimization of chance constrained problems. We assume that the probability measure with which the constraints are evaluated is known only through its moments. The algorithm proceeds in two phases. In the first phase the probability distribution is (coarsely) discretized and solved to global optimality using a stochastic algorithm. We only assume that the stochastic algorithm exhibits a weak* convergence to a probability measure assigning all its mass to the discretized problem. A diffusion process is derived that has this convergence property. In the second phase, the discretization is improved by solving another nonlinear programming problem. It is shown that the algorithm converges to the solution of the original problem. We discuss the numerical performance of the algorithm and its application to process design.
“…To obtain Markov chain annealing algorithms we simply replace the fixed temperature T in the above Markov chain sampling methods by a temperature schedule {Tk} (where typically Tk --0). We can establish a weak convergence result for a nonstationary continuous (4.13) We remark that there has been a lot of work establishing convergence results for discrete state Markov chain annealing algorithms [6], [24]- [27], and also for the Markov diffusion annealing algorithm [7], [28], [29]. However, there are very few convergence results for continuous state Markov chain algorithms.…”
“…The constant Co) plays a critical role in the convergence of Xk as k --+ oo and also Y(t) as t -+ oo. In [28] it is shown that the constant C( (denoted there by co) has an interpretation in terms of the action functional for a family of perturbed dynamical systems; see [28] for a further discussion of C) including some examples. 2.…”
Section: Y(tk1l) N Y(tk) -(Tk+ -Tk)vu(y(tk)) + C(tk)(w(tk+) -W(tk)) -mentioning
confidence: 99%
“…The asymptotic behavior of Y(t) as t --oo has been studied intensively by a number of researchers. In [7], [29] convergence results where obtained by considering a version of (4.17) with a reflecting boundary; in [28] the reflecting boundary was removed. Our analysis of {Xk} is based on the analysis of Y(t) developed in [28] where the following result is proved: if U(.)…”
We study the Langevin algorithm on C" n-dimensional compact connected Riemannian manifolds and on R", allowing the energy function U to vary with time. We find conditions under which the distribution of the process at hand becomes indistinguishable as t + co from the "instantaneous" equilibrium distribution. Such conditions do not necessarily imply that U ( t ) converges pointwise as t + 00.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.