Variational estimates for martingale transforms Pavel Zorin-Kranich University of Bonn 2020-06-09 Joint work with P. Friz. Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms 1/ 9
Rough paths Defjnition Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms . 1 / p |π u l β 1 , u l | p ) l = 1 β l max ( l max , u 0 <β―< u lmax sup V p π = , 1 / p l = 1 β l max A p -rough path, 2 < p < 3, is a pair X βΆ [ 0 , β) β H , π βΆ Ξ = {( s , t ) | 0 β€ s < t < β} β H β H such that X β V p loc , π β V p / 2 2/ 9 (Chenβs relation) p-variation: V p X = sup l max , u 0 <β―< u lmax ( loc , and for s < t < u π s , u = π s , t + π t , u + ( X u β X t ) β ( X t β X s ). | X u l β X u l β 1 | p ) How to check the conditions X β V p and π β V p / 2 ?
Rough path lifts of martingales Let M = ( M t ) be a (Hilbert space valued) cΓ dlΓ g martingale. Let Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms How to incorporate e.g. fractional Brownian motion? Q: what is the appropriate generality for these lifting results? There exist rough path lifts of over processes, e.g. LΓ©vy processes. 3/ 9 ( s , t ] Then, a.s., the pair ( M , π ) is a p -rough path for any p > 2. π s , t βΆ= β« ( M u β β M s ) β dM u . βΆ Chenβs relation β from ItΓ΄ integration βΆ Bound for V p M : LΓ©pingle 1976. βΆ Bounds for V p / 2 π : βΆ M Brownian motion: Lyons 1998 βΆ M has continuous paths: Friz+Victoir 2006 βΆ M dyadic: Do+Muscalu+Thiele 2010, βΆ M has cΓ dlΓ g paths: Chevyrev+Friz 2017, KovaΔ+ZK 2018.
Joint rough path lifts All martingales and processes are adapted, cΓ dlΓ g, Hilbert space valued. Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms The proof also recovers existence of ItΓ΄ integrals and estimates for New in this result: Motivation: dY = a ( Y ) d X + b ( Y ) dM is a p-rough path. ) 4/ 9 π ( X p-rough path ( 2 < p < 3 ). Then, a.s., the pair of processes Let M = ( M t ) be a cΓ dlΓ g martingale and ( X , π) a deterministic cΓ dlΓ g Theorem (Friz+ZK 2020+) β« X u β β dM u M ) , ( β« M u β β dX u β« M u β β dM u βΆ Variational estimates for ItΓ΄ integrals β« X dM , βΆ existence of and estimates for integrals β« M dX . π = β« M dM from previous slide.
Martingale transforms π Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms RHS is β² β Sf β L q 0 β Sg β L q 0 . In this case, any r > 1 works. The supremum is taken over increasing sequences of stopping times π = (π k ) . β L q 1 β Sg β L q 0 . 1 / p β π k β 1 < j β€π k sup ( k β(β β 5/ 9 β Theorem (Main estimate) Martingale in t variable, discrete version of area integral. β V r Ξ ( f , g )β s < j β€ t Let ( f n ) n ββ be a discrete time adapted process and ( g n ) n ββ a discrete time martingale. Defjne paraproduct Ξ s , t ( f , g ) βΆ= β ( f j β 1 β f s ) dg j , dg j = g j β g j β 1 . Let 1 β€ p β€ β , 0 < q 1 β€ β , 1 β€ q 0 < β . Defjne q by 1 / q = 1 / q 0 + 1 / q 1 and suppose 1 / r < 1 / 2 + 1 / p. Then, with β g β L q = (π½| g | q ) 1 / q , Sg = [ g ] 1 / 2 , β L q β² sup | f j β 1 β f π k β 1 |) p ) βΆ If f is a martingale, p = 2, 1 β€ q 1 < β , then by BDG inequality the βΆ For general f , RHS is β€ β V p f β L q 0 β Sg β L q 0 and r = p / 2 works.
Discrete approximation of adapted processes lim Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms Given π > 0, consider the adapted partition Proof. p β ( p , β) βͺ {β} . for any Μ p ) in Defjnition 6/ 9 If f β L q ( V p ) for some q > 0 and p > 1 , then Lemma βΆ= f β t ,πβ . t f (π) β t , πβ βΆ= max { s β π | s β€ t }, For an adapted partition π , let An adapted partition π = (π j ) j is an increasing sequence of stopping times. Adapted partitions are ordered by a.s. inclusion of the sets {π j | j β β} . The set of adapted partitions is directed, so lim π makes sense. π f (π) = f L q ( V Μ π j + 1 (π) βΆ= inf { t > π j (π) | π 0 βΆ= 0 , | | f t β f π j (π) |(π) > π}.
Stopping time reduction β Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms where the supremum is taken over all adapted partitions π . (1) β L q , 1 /π β π ) |Ξ t , t β² |) π j β 1 β€ t < t β² β€π j sup ( j = 1 β β( f adapted process, g martingale β π Then, for every 0 < π < r < β and q β ( 0 , β] , we have Lemma Suppose 1 / r < 1 / p + 1 / 2 . Then Theorem (Main estimate) Square function: Sg = [ g ] 1 / 2 , 7/ 9 Martingale transform: Ξ s , t ( f , g ) = β s < j β€ t ( f j β 1 β f s ) dg j HΓΆlder exponents: 1 / q = 1 / q 0 + 1 / q 1 . β V r Ξ β L q (β¦) β² β V p f β L q 1 (β¦) β Sg β L q 0 (β¦) . The V r norm is estimated as follows. Let (Ξ s , t ) s β€ t be a cΓ dlΓ g adapted sequence with Ξ t , t = 0 for all t. β V r Ξ β L q β² sup
LΓ©pingleβs inequality Above stopping time argument fjrst used in the following result. Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms inequalities that follow from weighted inequalities by OsΘ©kowski. For dealing with martingale transforms, we use vector-valued BDG Weighted inequalities imply vector-valued inequalities. Classical LΓ©pingleβs inequality is the case w β‘ 1, A p ( w ) = 1. p β 1 β L β ( w ) π ) βπ½( w | β± π stopping time sup A p ( w ) βΆ= For 1 < p < β and 2 < r, we have Let M be a martingale and w a positive random variable. Theorem (ZK 2019) 8/ 9 β V r M β L p ( w ) β€ C p , r A p ( w ) max ( 1 , 1 /( p β 1 )) β M β L p ( w ) , where the A p charactersitic is given by π )π½( w β 1 /( p β 1 ) | β±
Integration by parts β L q β( β β j = 1 ( β π j β 1 < u β€π j |Ξ X u Ξ u M | 2 ) π/ 2 ) 1 /π β HΓΆlder β² β€ β( β β j = 1 β π j β 1 < u β€π j |Ξ u M | 2 ) 1 / 2 β β L q Pavel Zorin-Kranich (U Bonn) β Variational estimates for martingale transforms β vector BDG ( X , π) rough path, M martingale β We do this by partial integration: Ξ ( M , X ) βΆ= π M π X β Ξ ( X , M ) β π[ X , M ]. The bracket is given by u β€ T Ξ X u Ξ M u , Variation norm estimate for the bracket: β V r [ X , M ]β L q β L q β² stopping β( β β j = 1 ( sup π j β 1 < t < t β² β€π j |π[ X , M ] t , t β² |) π ) 1 /π β 9/ 9 So far we have estimated β« X dM and β« M dM . Next, we want to construct and estimate Ξ ( M , X ) = β« M dX . [ X , M ] T = β Ξ M u = M u β M u β . V p X β β
Recommend
More recommend