Particle Methods for Rare Event Monte Carlo Paul Dupuis Division of Applied Mathematics Brown University (Thomas Dean (Oxford)) ASEAS, Arlington, VA March, 2009 Paul Dupuis (Brown University) March, 2009
Particle methods for rare event Monte Carlo The use of branching processes to estimate small probabilities Paul Dupuis (Brown University) March, 2009
Particle methods for rare event Monte Carlo The use of branching processes to estimate small probabilities Summary: Paul Dupuis (Brown University) March, 2009
Particle methods for rare event Monte Carlo The use of branching processes to estimate small probabilities Summary: The design of such schemes was (until recently) poorly understood. Paul Dupuis (Brown University) March, 2009
Particle methods for rare event Monte Carlo The use of branching processes to estimate small probabilities Summary: The design of such schemes was (until recently) poorly understood. Design should be based on subsolutions to an associated HJB equation. Paul Dupuis (Brown University) March, 2009
Particle methods for rare event Monte Carlo The use of branching processes to estimate small probabilities Summary: The design of such schemes was (until recently) poorly understood. Design should be based on subsolutions to an associated HJB equation. Obtain necessary and su¢cient conditions for asymptotically optimal performance. Paul Dupuis (Brown University) March, 2009
Example: A tandem queue with server slow-down (Ethernet control) Paul Dupuis (Brown University) March, 2009
Example: A tandem queue with server slow-down (Ethernet control) Paul Dupuis (Brown University) March, 2009
Example: A tandem queue with server slow-down (Ethernet control) p n = P f Q 2 exceeds n before Q = ( 0 ; 0 ) j Q ( 0 ) = ( 1 ; 0 ) g : Paul Dupuis (Brown University) March, 2009
Example: A tandem queue with server slow-down (Ethernet control) p n = P f Q 2 exceeds n before Q = ( 0 ; 0 ) j Q ( 0 ) = ( 1 ; 0 ) g : Also, analogous non-Markovian model. Paul Dupuis (Brown University) March, 2009
Model problem and large deviation scaling As a general Markov model one can consider iid random vector …elds � v i ( y ) ; y 2 R d � , and the process i + 1 X n i + 1 = X n nv i ( X n X n i ) ; 0 = x : Paul Dupuis (Brown University) March, 2009
Model problem and large deviation scaling As a general Markov model one can consider iid random vector …elds � v i ( y ) ; y 2 R d � , and the process i + 1 X n i + 1 = X n nv i ( X n X n i ) ; 0 = x : De…ne H ( y ; � ) = log E exp h �; v i ( y ) i ; L ( y ; � ) = sup � 2 R d [ h �; � i � H ( y ; � )] X n ( i = n ) = X n i ; piecewise linear interpolation for t 6 = i = n : Paul Dupuis (Brown University) March, 2009
Model problem and large deviation scaling (cont’d) Under conditions f X n ( � ) g satis…es a Large Deviation Principle with rate function Z T L ( �; _ I T ( � ) = � ) dt 0 if � is AC and � ( 0 ) = x , and I T ( � ) = 1 else. Paul Dupuis (Brown University) March, 2009
Model problem and large deviation scaling (cont’d) Under conditions f X n ( � ) g satis…es a Large Deviation Principle with rate function Z T L ( �; _ I T ( � ) = � ) dt 0 if � is AC and � ( 0 ) = x , and I T ( � ) = 1 else. Heuristically, for T < 1 , given � , small � > 0 and large n ( ) k X n ( t ) � � ( t ) k � � � e � nI T ( � ) : P sup 0 � t � T Paul Dupuis (Brown University) March, 2009
Model problem and large deviation scaling (cont’d) Let C = f trajectories that hit B prior to A g . To estimate: p n ( x ) = P f X n 2 C j X n ( 0 ) = x g : Paul Dupuis (Brown University) March, 2009
Model problem and large deviation scaling (cont’d) Under mild conditions: � 1 n log p n ( x ) ! inf f I T ( � ) : � enters B prior to A before T ; T < 1g = � ( x ) : Paul Dupuis (Brown University) March, 2009
Some estimation generalities For standard Monte Carlo we average iid copies of 1 f X n 2 C g . One 1 needs K � e n � samples for bounded relative error [std dev/ p n ( x ) ]. Paul Dupuis (Brown University) March, 2009
Some estimation generalities For standard Monte Carlo we average iid copies of 1 f X n 2 C g . One 1 needs K � e n � samples for bounded relative error [std dev/ p n ( x ) ]. Alternative approach: construct iid random variables � n 1 ; : : : ; � n K with 2 E � n 1 = p n ( x ) and use the unbiased estimator = � n 1 + � � � + � n q n ; K ( x ) : K ^ : K Paul Dupuis (Brown University) March, 2009
Some estimation generalities For standard Monte Carlo we average iid copies of 1 f X n 2 C g . One 1 needs K � e n � samples for bounded relative error [std dev/ p n ( x ) ]. Alternative approach: construct iid random variables � n 1 ; : : : ; � n K with 2 E � n 1 = p n ( x ) and use the unbiased estimator = � n 1 + � � � + � n q n ; K ( x ) : K ^ : K Performance determined by variance of � n 1 , and since unbiased by 3 1 ) 2 . E ( � n Paul Dupuis (Brown University) March, 2009
Some estimation generalities For standard Monte Carlo we average iid copies of 1 f X n 2 C g . One 1 needs K � e n � samples for bounded relative error [std dev/ p n ( x ) ]. Alternative approach: construct iid random variables � n 1 ; : : : ; � n K with 2 E � n 1 = p n ( x ) and use the unbiased estimator = � n 1 + � � � + � n q n ; K ( x ) : K ^ : K Performance determined by variance of � n 1 , and since unbiased by 3 1 ) 2 . E ( � n By Jensen’s inequality 4 � 1 1 ) 2 � � 2 1 = � 2 n log E ( � n n log E � n n log p n ( x ) ! 2 � ( x ) : Paul Dupuis (Brown University) March, 2009
Some estimation generalities For standard Monte Carlo we average iid copies of 1 f X n 2 C g . One 1 needs K � e n � samples for bounded relative error [std dev/ p n ( x ) ]. Alternative approach: construct iid random variables � n 1 ; : : : ; � n K with 2 E � n 1 = p n ( x ) and use the unbiased estimator = � n 1 + � � � + � n q n ; K ( x ) : K ^ : K Performance determined by variance of � n 1 , and since unbiased by 3 1 ) 2 . E ( � n By Jensen’s inequality 4 � 1 1 ) 2 � � 2 1 = � 2 n log E ( � n n log E � n n log p n ( x ) ! 2 � ( x ) : An estimator is called asymptotically e¢cient if 5 n !1 � 1 1 ) 2 � 2 � ( x ) : n log E ( � n lim inf Paul Dupuis (Brown University) March, 2009
Splitting type schemes Pure branching methods (also called multi-level splitting) Paul Dupuis (Brown University) March, 2009
Splitting type schemes Pure branching methods (also called multi-level splitting) Branching with killing [RESTART, DPR] Paul Dupuis (Brown University) March, 2009
Splitting type schemes Pure branching methods (also called multi-level splitting) Branching with killing [RESTART, DPR] Interacting particle systems (Del Moral et. al.) Paul Dupuis (Brown University) March, 2009
Construction of splitting estimators Pure branching. A certain number [proportional to n ] of splitting thresholds C n r are de…ned which enhance migration, e.g., Paul Dupuis (Brown University) March, 2009
Construction of splitting estimators Pure branching. A certain number [proportional to n ] of splitting thresholds C n r are de…ned which enhance migration, e.g., A single particle is started at x that follows the same law as X n , but branches into a number of independent copies each time a new level is reached. Paul Dupuis (Brown University) March, 2009
Construction of splitting estimators (cont’d) Paul Dupuis (Brown University) March, 2009
Construction of splitting estimators (cont’d) The number of new particles M can be random (though independent of past data), and a multiplicative weight w i is assigned to the i th descendent, where M X E w i = 1 : i = 1 Paul Dupuis (Brown University) March, 2009
Construction of splitting estimators (cont’d) Evolution continues until every particle has reached either A or B . Let M n = total number of particles generated x X n j ( t ) = trajectory of j th particle, W n = product of weights assigned to j along path j Paul Dupuis (Brown University) March, 2009
Construction of splitting estimators (cont’d) Evolution continues until every particle has reached either A or B . Let M n = total number of particles generated x X n j ( t ) = trajectory of j th particle, W n = product of weights assigned to j along path j Then M n X x � n = j 2 C g W n 1 f X n j j = 1 Paul Dupuis (Brown University) March, 2009
Subsolutions for branching processes Now consider the asymptotic rate of decay as a function of y : n !1 � 1 � ( y ) = lim n log p n ( y ) = inf f I T ( � ) : � ( 0 ) = y ; � enters B prior to A before T ; T < 1g : Paul Dupuis (Brown University) March, 2009
Subsolutions for branching processes Now consider the asymptotic rate of decay as a function of y : n !1 � 1 � ( y ) = lim n log p n ( y ) = inf f I T ( � ) : � ( 0 ) = y ; � enters B prior to A before T ; T < 1g : Let H ( y ; � ) = � H ( y ; � � ) [recall H ( y ; � ) = log E exp h �; v i ( y ) i ] : Paul Dupuis (Brown University) March, 2009
Subsolutions for branching processes (cont’d) � ( y ) is a weak-sense solution to the PDE Paul Dupuis (Brown University) March, 2009
Recommend
More recommend