Characterization of multi parameter BMO spaces through commutators Stefanie Petermichl Université Paul Sabatier IWOTA Chemnitz August 2017 S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 1 / 30
history Hankel vs. Toeplitz on T P ± projection operator onto non-negative and negative frequencies. A Hankel operator with symbol b is H b : L 2 + H 2, f P bp + f b BMO characterises boundedness. A Toeplitz operator with symbol b is b L characterises boundedness. T b : L 2 + H 2 +, f P + bp + f S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 2 / 30
history Hankel Operators P bp + H b = sup g H 2 =1 sup f H 2 =1 (H b f, g) = sup g H 2 =1 sup f H 2 =1 (P (bf ), g) = sup g H 2 =1 sup f H 2 + =1 (P bf, g) = sup g H 2 =1 sup f H 2 =1 (P b, f g) Anti-analytic part of b defines bounded linear functional on H 1 L 1. Extend by Hahn Banach to all of L 1, i.e. a bounded function with the same anti-analytic part as b. Using H 1 BMO duality, we get the characterisation of BMO. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 3 / 30
history Toeplitz operators P + bp + Clearly T b b But L also characterises boundedness: It is easy to see that λ n P + λ n I in L 2 in SOT. Nothing happens to such f with FS cut off at n. So λ n P + bp + λ n b in L 2 in SOT. Now as multiplication operators: b sup n λ n P + bp + λ n P + bp + S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 4 / 30
Commutators with the Hilbert transform Commutators [H, b] [b, H]f = b Hf H(bf ) where H is the Hilbert transform and b is multiplication by the function b. If we write H = P + P and I = P + + P then [b, H] = [(P + + P )b, (P + P )] = 2P bp + 2P + bp, two Hankel operators with orthogonal ranges. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 5 / 30
Commutators with the Hilbert transform Extensions of [H, b] Riesz transform commutators and similar. Coifman, Rochberg, Weiss, Uchiyama, Lacey... The passage to several parameters, initiation. Cotlar, Ferguson, Sadosky... Thiele, Muscalu, Tao, Journe, Holmes, Lacey, Pipher, Strouse, Wick, P.... S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 6 / 30
Commutators with the Hilbert transform Commutators [H 1 H 2, b] with b(x 1, x 2 ) Tensor product one-parameter case. Theorem (Ferguson, Sadosky) [H 1 H 2, b] bounded in L 2 iff b bmo little BMO b bmo = max{sup b(x 1, ) BMO, sup b(, x 2 ) BMO } x 1 x 2 i.e. b uniformly in BMO in each variable separately or of bounded mean oscillation on rectangles. more Hilbert transforms [H 1 H 2 H 3, b] etc implicit. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 7 / 30
Commutators with the Hilbert transform Commutators [H 1, [H 2, b]] with b(x 1, x 2 ) Theorem (Ferguson, Lacey) [H 1, [H 2, b]] bounded in L 2 iff b BMO product BMO b 2 BMO = sup O 1 O (b, h R ) 2 R O more iterations [H 1, [H 2, [H 3, b]]] not implicit, but Terwilliger, Lacey. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 8 / 30
Commutators with the Hilbert transform Lower estimates for Hilbert commutators Ferguson-Sadosky: elegant soft argument, based on Toeplitz forms. Ferguson-Lacey: extremely technical hard real analysis argument based on Hankel forms. Using scale analysis, Schwarz tail estimates, geometric facts on distribution of rectangles in the plane... S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 9 / 30
Commutators with the Hilbert transform Commutators [H 2, [H 3 H 1, b]] with b(x 1, x 2, x 3 ) Theorem (Ou, Strouse, P.) [H 2, [H 3 H 1, b]] bounded in L 2 iff b BMO (13)2 little product BMO b BMO(13)2 = max{sup b(x 1,, ) BMO, sup b(,, x 3 ) BMO } x 1 x 3 b uniformly in product BMO when fixing variables x 3 and x 1. more Hilbert transforms and more iterations implicit. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 10 / 30
Commutators with the Hilbert transform Commutators [H 2, [H 3 H 1, b]] with b(x 1, x 2, x 3 ) Infact TFAE: 1 b BMO (13)2 2 [H 2, [H 1, b]] and [H 2, [H 3, b]] bounded in L 2 (T 3 ) 3 [H 2, [H 3 H 1, b]] bounded in L 2 (T 3 ). 1 eq 2: Wiener s theorem and Ferguson, Lacey: [H 2, [H 1, b]]f (x 1, x 2 )g(x 3 ) = g(x 3 )[H 2, [H 1, b]]f (x 1, x 2 ) 2 eq 3: Toeplitz argument: typical terms that arise: P + 1 P+ 2 bp 1 P 2 and P+ 1 P+ 2 P+ 3 bp 1 P 2 P+ 3 S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 11 / 30
Riesz Riesz commutators and the absence of Hankel and Toeplitz. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 12 / 30
One parameter: [R i, b] It is a classical result by Coifman, Rochberg and Weiss, that the Riesz transform commutators classify BMO. For each symbol b BMO we may choose the worst Riesz transform. In this sense But Testing class for CZOs. [b, R i ] 2 2 b BMO b BMO sup [b, R i ] 2 2 i S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 13 / 30
One parameter tensor product: [R 1,i1 R 2,i2, b] Through use of the little BMO norm, one sees that [b, R 1,i1 R 2,i2 ] 2 2 b bmo Through a direct calculation using the little BMO norm one also sees b bmo sup i 1,i 2 [b, R 1,i1 R 2,i2 ] 2 2 S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 14 / 30
multi-parameter: [R 2,j2, [R 1,j1, b]] Theorem (Lacey, Pipher, Wick, P.) sup [R 2,j2, [R 1,j1, b]] b BMO. j 1,j 2 By BMO, we mean Chang Fefferman product BMO. Implicit generalizations with similar proof. Testing for CZOs. This means we test a symbol on Riesz transforms. The estimate then self improves to all operators of the same type as Riesz transforms. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 15 / 30
multi-parameter tensor product: [R 2,j2, [R 1,j1 R 3,j3, b]] Theorem (Ou, Strouse, P.) where we mean little product BMO. b BMO(13)2 [R 2,j2, [R 1,j1 R 3,j3, b]] Implicit generalizations but proof is substantially more difficult when dimensions are greater than 2 or when slots contain tensor products of more than 2. Testing for (paraproduct-free) Journé operators - we will see later what these are. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 16 / 30
Cones Riesz transforms do not have the same relation to projections as the Hilbert transform does. Replace by well chosen, smooth half plane projections that are CZO. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 17 / 30
Cones Depending on the make of the symbol function, a multi parameter skeleton of cones of large aperture is chosen via a probabilistic procedure. The others are filled in using tiny cones via a Toeplitz argument. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 18 / 30
Polynomials We have now showed that there is a lower estimate if one can choose from cone operators. How does this help for Riesz transforms? Observe: [T 1 T 2, b] = T 1 [T 2, b] + [T 1, b]t 2 If the commutator with T 1 T 2 is large, then one of the commutators with T i has to be large. Riesz transforms have Fourier symbols ξ i on S n (monomials) well adapted for polynomial approximation. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 19 / 30
Passage to tensor products of Riesz transforms What goes wrong: If [R 2 1,i 1 R 2,i2, b] large, cannot say [R 1,i1 R 2,i2, b] remains large. It is too wasteful to just have any lower estimates of tensor products of cone operators. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 20 / 30
Passage to tensor products of Riesz transforms These strip operators work well on products of S 1 and a deep generalization using a probabilistic construction and zonal harmonics will work on higher dimensional spheres. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 21 / 30
Passage to tensor products of Riesz transforms Looking like this: S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 22 / 30
Passage to tensor products of Riesz transforms and this: S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 23 / 30
Passage to tensor products of Riesz transforms and this: S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 24 / 30
Upper estimates for Hilbert commutators none of them are difficult (for the Hilbert transform), there are arguments using simple operator theory, but here is the proof that will extend. This simple operator is useful S : h I h I h I+ The indexed intervals are dyadic intervals and h I denotes the Haar basis. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 25 / 30
The dyadic Hilbert transform H gives access to harmonic conjugates, z z is analytic in D with boundary values e it = cos(t) + i sin(t). So H(sin) = cos. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 26 / 30
The dyadic Hilbert transform The Hilbert transform can be written as expectation of dyadic Hilbert transforms. The proof is very elementary and explicit. It turns out this is the right tool to capture cancellation in a commutator. This tool was invented to address a question on commutators raised by Pisier. Its thrust lies in the immediate reduction of commutator bounds to so-called paraproducts. These are triangular sums in the naive multiplication ( (f, h I )h I ) ( (b, h J )h J ) S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 27 / 30
Upper estimates For the Riesz case need a stability estimate for Journe commutators to make the argument work: [J 1,..., [J t, b],..., ] C b with C 0 when defining constants of J do. Use very general Haar shift operators for Journe operators that locally look like tensor products. (Hytonen, Martikainen) S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 28 / 30
Journé Operators T defined on testing class C0 C0 with two CZO kernels K 1, K 2 such that with f 1 (y 1 ), g 1 (x 1 ), f 2 (y 2 )g 2 (x 2 ) (T (f 1 f 2 ), g 1 g 2 ) = f 1 (y 1 ) K 1 (x 1, y 1 )f 2, g 2 g 1 (x 1 )dx 1 dy 1 if f 1, g 1 disjoint support (T (f 1 f 2 ), g 1 g 2 ) = f 2 (y 2 ) K 2 (x 2, y 2 )f 1, g 1 g 2 (x 2 )dx 2 dy 2 if f 2, g 2 disjoint support weak boundedness property (T (1 I 1 J ), 1 I 1 J ) I J Apply T to test function f 1 using first set of variables of the kernel. Get another CZO (depending on f 1 ) then applied to f 2. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 29 / 30
αi1,j 1,K 1,I 2,J 2,K 2 (f, h I1 h I2 )h J1 h J2 Representation Such operators have a representation using terms of the form where the dyadic size difference of I i and J i to K i is constant for each constellation. Then one sums in this difference and averages. There is a decay with this distance of the coefficients. If there is a tensor product of CZO the coefficients split, but for true Journe operators, they are sticky. The matching mixed BMO condition is exactly correct to estimate commutators for these. S. Petermichl (Université Paul Sabatier) Commutators and BMO Chemnitz 30 / 30