Multilevel Markov Chain Monte Carlo for Uncertainty Quantification in Subsurface Flow
Christian Ketelsen
Applied Mathematics,听
Date and time:聽
Friday, January 31, 2014 - 3:00pm
尝辞肠补迟颈辞苍:听
ECCR 265
础产蝉迟谤补肠迟:听
In this talk we address the problem of the intractably large computational cost of standard Markov chain Monte Carlo (MCMC) methods for quantifying uncertainty in subsurface flow simulations. The multilevel Monte Carlo method has been shown to be an effective variance reduction technique for accelerating the simulation when the model parameters can be represented by a simple prior distribution. In state-of-the-art subsurface simulation the model parameters must be conditioned on observe physical data. Sampling from this complicated distribution is carried out by the Markov chain Monte Carlo method. In this talk we extend the multilevel Monte Carlo methodology to the Markov chain Monte Carlo setting and demonstrate the effectiveness of the method via a model problem of single-phase flow in a random medium.