Localization and projections on bi–parameter BMO R. Lechner P. F. X. Müller J. Kepler University, Linz Warwick, June, 2015 Overview 1 A description of the problem class 2 One dimension 3 Two dimensions Overview 1 A description of the problem class 2 One dimension 3 Two dimensions A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. A description of the problem class • Let X be a Banach space • T : X → X a linear operator Find conditions on T and X such that the identity on X factors through T (or Id −T ), i.e. X E Id P X /X O T kEkkP k ≤ C. /X • The problem has finite dimensional (quantitative) and infinite dimensional (qualitative) aspects. • Classical examples include: `p (Pelczynski), C([0, 1]) (Lindenstrauss-Pelczynski), Lp (Gamlen-Gaudet), L1 (Enflo-Starbird), `pn (Bourgain-Tzafriri). • Generically, we expect that T has a large diagonal with respect to an unconditional basis in X. • We consider special cases where X = H p (H q ) with the bi-parameter Haar system as its unconditional basis. Overview 1 A description of the problem class 2 One dimension 3 Two dimensions Dyadic H p k • D = {[ k−1 2n , 2n [ : k ≥ 0, n ≥ 0} denotes the dyadic intervals on the unit interval, • hI the L∞ -normalized Haar function, I ∈ D. • Let f = P I∈D aI hI be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X a2I h2I I∈D 1/2 . • The norm of the one-parameter Hardy space H p , 1 ≤ p < ∞ is defined by kf k Hp = k S(f )k Lp = Z 0 1 X I∈D a2I h2I (x) p/2 dx 1/p . Dyadic H p k • D = {[ k−1 2n , 2n [ : k ≥ 0, n ≥ 0} denotes the dyadic intervals on the unit interval, • hI the L∞ -normalized Haar function, I ∈ D. • Let f = P I∈D aI hI be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X a2I h2I I∈D 1/2 . • The norm of the one-parameter Hardy space H p , 1 ≤ p < ∞ is defined by kf k Hp = k S(f )k Lp = Z 0 1 X I∈D a2I h2I (x) p/2 dx 1/p . Dyadic H p k • D = {[ k−1 2n , 2n [ : k ≥ 0, n ≥ 0} denotes the dyadic intervals on the unit interval, • hI the L∞ -normalized Haar function, I ∈ D. • Let f = P I∈D aI hI be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X a2I h2I I∈D 1/2 . • The norm of the one-parameter Hardy space H p , 1 ≤ p < ∞ is defined by kf k Hp = k S(f )k Lp = Z 0 1 X I∈D a2I h2I (x) p/2 dx 1/p . Dyadic H p k • D = {[ k−1 2n , 2n [ : k ≥ 0, n ≥ 0} denotes the dyadic intervals on the unit interval, • hI the L∞ -normalized Haar function, I ∈ D. • Let f = P I∈D aI hI be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X a2I h2I I∈D 1/2 . • The norm of the one-parameter Hardy space H p , 1 ≤ p < ∞ is defined by kf k Hp = k S(f )k Lp = Z 0 1 X I∈D a2I h2I (x) p/2 dx 1/p . Dyadic H p k • D = {[ k−1 2n , 2n [ : k ≥ 0, n ≥ 0} denotes the dyadic intervals on the unit interval, • hI the L∞ -normalized Haar function, I ∈ D. • Let f = P I∈D aI hI be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X a2I h2I I∈D 1/2 . • The norm of the one-parameter Hardy space H p , 1 ≤ p < ∞ is defined by kf k Hp = k S(f )k Lp = Z 0 1 X I∈D a2I h2I (x) p/2 dx 1/p . Andrew 1D Theorem Let 1 < p < ∞, δ > 0 and T : H p → H p be a linear operator with large diagonal, i.e. hT hI , hI i ≥ δ|I|. Then we have Hp E Id kEkkP k ≤ C/δ, P Hp / Hp O T / Hp where the constant C > 0 is universal. By Gamlen-Gaudet construction (BI ) and a random choice of signs εI there P (ε) exists a block basis bI = K∈BI εK hK of the Haar system such that (ε) (ε) T bI = αI bI + small error, The orthogonal projection Qf = (Gamlen-Gaudet). P (ε) hf,bI i I∈D kb(ε) k2 bI 2 I αI ≥ δ. is bounded on H p Andrew 1D Theorem Let 1 < p < ∞, δ > 0 and T : H p → H p be a linear operator with large diagonal, i.e. hT hI , hI i ≥ δ|I|. Then we have Hp E Id kEkkP k ≤ C/δ, P Hp / Hp O T / Hp where the constant C > 0 is universal. By Gamlen-Gaudet construction (BI ) and a random choice of signs εI there P (ε) exists a block basis bI = K∈BI εK hK of the Haar system such that (ε) (ε) T bI = αI bI + small error, The orthogonal projection Qf = (Gamlen-Gaudet). P (ε) hf,bI i I∈D kb(ε) k2 bI 2 I αI ≥ δ. is bounded on H p Andrew 1D Theorem Let 1 < p < ∞, δ > 0 and T : H p → H p be a linear operator with large diagonal, i.e. hT hI , hI i ≥ δ|I|. Then we have Hp E Id kEkkP k ≤ C/δ, P Hp / Hp O T / Hp where the constant C > 0 is universal. By Gamlen-Gaudet construction (BI ) and a random choice of signs εI there P (ε) exists a block basis bI = K∈BI εK hK of the Haar system such that (ε) (ε) T bI = αI bI + small error, The orthogonal projection Qf = (Gamlen-Gaudet). P (ε) hf,bI i I∈D kb(ε) k2 bI 2 I αI ≥ δ. is bounded on H p Andrew 1D Theorem Let 1 < p < ∞, δ > 0 and T : H p → H p be a linear operator with large diagonal, i.e. hT hI , hI i ≥ δ|I|. Then we have Hp E Id kEkkP k ≤ C/δ, P Hp / Hp O T / Hp where the constant C > 0 is universal. By Gamlen-Gaudet construction (BI ) and a random choice of signs εI there P (ε) exists a block basis bI = K∈BI εK hK of the Haar system such that (ε) (ε) T bI = αI bI + small error, The orthogonal projection Qf = (Gamlen-Gaudet). P (ε) hf,bI i I∈D kb(ε) k2 bI 2 I αI ≥ δ. is bounded on H p Andrew 1D Theorem Let 1 < p < ∞, δ > 0 and T : H p → H p be a linear operator with large diagonal, i.e. hT hI , hI i ≥ δ|I|. Then we have Hp E Id kEkkP k ≤ C/δ, P Hp / Hp O T / Hp where the constant C > 0 is universal. By Gamlen-Gaudet construction (BI ) and a random choice of signs εI there P (ε) exists a block basis bI = K∈BI εK hK of the Haar system such that (ε) (ε) T bI = αI bI + small error, The orthogonal projection Qf = (Gamlen-Gaudet). P (ε) hf,bI i I∈D kb(ε) k2 bI 2 I αI ≥ δ. is bounded on H p The Gamlen-Gaudet construction The Gamlen-Gaudet construction The Gamlen-Gaudet construction The Gamlen-Gaudet construction The Gamlen-Gaudet construction Comments on Andrew 1D • Andrew precedes the study of operators with property A (Johnson-Maurey-Schechtman-Tzafriri). • The Rademacher system converges weakly to 0 in H p . • The operator T is preconditioned by multiplying the Haar system with highly oscillating Rademacher functions. (ε) (ε) • This gives that hT bI , bJ i ≈ 0, if I 6= J. • The second part consists of choosing signs εI such that (ε) (ε) (ε) hT bI , bI i ≥ δkbI k22 . (ε) • The block basis {bI : I ∈ D} is equivalent to the Haar system. • Andrew’s method of proof used a semi-random choice of signs εI , hence it is strictly limited to the range 1 < p < ∞. Comments on Andrew 1D • Andrew precedes the study of operators with property A (Johnson-Maurey-Schechtman-Tzafriri). • The Rademacher system converges weakly to 0 in H p . • The operator T is preconditioned by multiplying the Haar system with highly oscillating Rademacher functions. (ε) (ε) • This gives that hT bI , bJ i ≈ 0, if I 6= J. • The second part consists of choosing signs εI such that (ε) (ε) (ε) hT bI , bI i ≥ δkbI k22 . (ε) • The block basis {bI : I ∈ D} is equivalent to the Haar system. • Andrew’s method of proof used a semi-random choice of signs εI , hence it is strictly limited to the range 1 < p < ∞. Comments on Andrew 1D • Andrew precedes the study of operators with property A (Johnson-Maurey-Schechtman-Tzafriri). • The Rademacher system converges weakly to 0 in H p . • The operator T is preconditioned by multiplying the Haar system with highly oscillating Rademacher functions. (ε) (ε) • This gives that hT bI , bJ i ≈ 0, if I 6= J. • The second part consists of choosing signs εI such that (ε) (ε) (ε) hT bI , bI i ≥ δkbI k22 . (ε) • The block basis {bI : I ∈ D} is equivalent to the Haar system. • Andrew’s method of proof used a semi-random choice of signs εI , hence it is strictly limited to the range 1 < p < ∞. Comments on Andrew 1D • Andrew precedes the study of operators with property A (Johnson-Maurey-Schechtman-Tzafriri). • The Rademacher system converges weakly to 0 in H p . • The operator T is preconditioned by multiplying the Haar system with highly oscillating Rademacher functions. (ε) (ε) • This gives that hT bI , bJ i ≈ 0, if I 6= J. • The second part consists of choosing signs εI such that (ε) (ε) (ε) hT bI , bI i ≥ δkbI k22 . (ε) • The block basis {bI : I ∈ D} is equivalent to the Haar system. • Andrew’s method of proof used a semi-random choice of signs εI , hence it is strictly limited to the range 1 < p < ∞. Comments on Andrew 1D • Andrew precedes the study of operators with property A (Johnson-Maurey-Schechtman-Tzafriri). • The Rademacher system converges weakly to 0 in H p . • The operator T is preconditioned by multiplying the Haar system with highly oscillating Rademacher functions. (ε) (ε) • This gives that hT bI , bJ i ≈ 0, if I 6= J. • The second part consists of choosing signs εI such that (ε) (ε) (ε) hT bI , bI i ≥ δkbI k22 . (ε) • The block basis {bI : I ∈ D} is equivalent to the Haar system. • Andrew’s method of proof used a semi-random choice of signs εI , hence it is strictly limited to the range 1 < p < ∞. Comments on Andrew 1D • Andrew precedes the study of operators with property A (Johnson-Maurey-Schechtman-Tzafriri). • The Rademacher system converges weakly to 0 in H p . • The operator T is preconditioned by multiplying the Haar system with highly oscillating Rademacher functions. (ε) (ε) • This gives that hT bI , bJ i ≈ 0, if I 6= J. • The second part consists of choosing signs εI such that (ε) (ε) (ε) hT bI , bI i ≥ δkbI k22 . (ε) • The block basis {bI : I ∈ D} is equivalent to the Haar system. • Andrew’s method of proof used a semi-random choice of signs εI , hence it is strictly limited to the range 1 < p < ∞. Comments on Andrew 1D • Andrew precedes the study of operators with property A (Johnson-Maurey-Schechtman-Tzafriri). • The Rademacher system converges weakly to 0 in H p . • The operator T is preconditioned by multiplying the Haar system with highly oscillating Rademacher functions. (ε) (ε) • This gives that hT bI , bJ i ≈ 0, if I 6= J. • The second part consists of choosing signs εI such that (ε) (ε) (ε) hT bI , bI i ≥ δkbI k22 . (ε) • The block basis {bI : I ∈ D} is equivalent to the Haar system. • Andrew’s method of proof used a semi-random choice of signs εI , hence it is strictly limited to the range 1 < p < ∞. Overview 1 A description of the problem class 2 One dimension 3 Two dimensions Mixed-norm Hardy spaces H p (H q ) • R = {I × J : I, J ∈ D} denotes the dyadic rectangles on the unit square, • hI×J (x, y) = hI (x)hJ (y) the L∞ -normalized tensor product Haar function, I × J ∈ R. P • Let f = I×J∈R aI×J hI×J be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X I×J∈R a2I×J h2I×J 1/2 . • we define the norm of the bi-parameter Hardy spaces H p (H q ), 1 ≤ p, q < ∞ by kf kH p (H q ) = Z 0 1Z 1 0 X I∈D a2I×J h2I×J (x, y) q/2 dy p/q dx 1/p . Mixed-norm Hardy spaces H p (H q ) • R = {I × J : I, J ∈ D} denotes the dyadic rectangles on the unit square, • hI×J (x, y) = hI (x)hJ (y) the L∞ -normalized tensor product Haar function, I × J ∈ R. P • Let f = I×J∈R aI×J hI×J be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X I×J∈R a2I×J h2I×J 1/2 . • we define the norm of the bi-parameter Hardy spaces H p (H q ), 1 ≤ p, q < ∞ by kf kH p (H q ) = Z 0 1Z 1 0 X I∈D a2I×J h2I×J (x, y) q/2 dy p/q dx 1/p . Mixed-norm Hardy spaces H p (H q ) • R = {I × J : I, J ∈ D} denotes the dyadic rectangles on the unit square, • hI×J (x, y) = hI (x)hJ (y) the L∞ -normalized tensor product Haar function, I × J ∈ R. P • Let f = I×J∈R aI×J hI×J be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X I×J∈R a2I×J h2I×J 1/2 . • we define the norm of the bi-parameter Hardy spaces H p (H q ), 1 ≤ p, q < ∞ by kf kH p (H q ) = Z 0 1Z 1 0 X I∈D a2I×J h2I×J (x, y) q/2 dy p/q dx 1/p . Mixed-norm Hardy spaces H p (H q ) • R = {I × J : I, J ∈ D} denotes the dyadic rectangles on the unit square, • hI×J (x, y) = hI (x)hJ (y) the L∞ -normalized tensor product Haar function, I × J ∈ R. P • Let f = I×J∈R aI×J hI×J be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X I×J∈R a2I×J h2I×J 1/2 . • we define the norm of the bi-parameter Hardy spaces H p (H q ), 1 ≤ p, q < ∞ by kf kH p (H q ) = Z 0 1Z 1 0 X I∈D q/2 p/q 1/p a2I×J h2I×J (x, y) dy dx . Mixed-norm Hardy spaces H p (H q ) • R = {I × J : I, J ∈ D} denotes the dyadic rectangles on the unit square, • hI×J (x, y) = hI (x)hJ (y) the L∞ -normalized tensor product Haar function, I × J ∈ R. P • Let f = I×J∈R aI×J hI×J be a finite linear combination, • then the square function S(f ) of f is given by S(f ) = X I×J∈R a2I×J h2I×J 1/2 . • we define the norm of the bi-parameter Hardy spaces H p (H q ), 1 ≤ p, q < ∞ by kf kH p (H q ) = Z 0 1Z 1 0 X I∈D q/2 p/q 1/p a2I×J h2I×J (x, y) dy dx . Capon’s theorem: H p (H q ) is primary (augmented) Theorem Let 1 < p, q < ∞ or p = q = 1. For any operator T the identity on H p (H q ) factors through H = T or H = Id −T , i.e. Id H p (H q ) E / H p (H q ) O P H p (H q ) H kEkkP k ≤ C, / H p (H q ) where C = C(kT k). Capon invents a specific bi-parameterPGamlen-Gaudet selection process which leads to a block basis bI×J = K×L∈BI×J hK×L such that HbI×J = αI×J bI×J + small error, and the projection Qf = P hf,bI×J i I×J∈R kbI×J k22 bI×J αI×J ≥ 1/2. is bounded on H p (H q ). Capon’s theorem: H p (H q ) is primary (augmented) Theorem Let 1 < p, q < ∞ or p = q = 1. For any operator T the identity on H p (H q ) factors through H = T or H = Id −T , i.e. Id H p (H q ) E / H p (H q ) O P H p (H q ) H kEkkP k ≤ C, / H p (H q ) where C = C(kT k). Capon invents a specific bi-parameterPGamlen-Gaudet selection process which leads to a block basis bI×J = K×L∈BI×J hK×L such that HbI×J = αI×J bI×J + small error, and the projection Qf = P hf,bI×J i I×J∈R kbI×J k22 bI×J αI×J ≥ 1/2. is bounded on H p (H q ). Capon’s theorem: H p (H q ) is primary (augmented) Theorem Let 1 < p, q < ∞ or p = q = 1. For any operator T the identity on H p (H q ) factors through H = T or H = Id −T , i.e. Id H p (H q ) E / H p (H q ) O P H p (H q ) H kEkkP k ≤ C, / H p (H q ) where C = C(kT k). Capon invents a specific bi-parameterPGamlen-Gaudet selection process which leads to a block basis bI×J = K×L∈BI×J hK×L such that HbI×J = αI×J bI×J + small error, and the projection Qf = P hf,bI×J i I×J∈R kbI×J k22 bI×J αI×J ≥ 1/2. is bounded on H p (H q ). Capon’s theorem: H p (H q ) is primary (augmented) Theorem Let 1 < p, q < ∞ or p = q = 1. For any operator T the identity on H p (H q ) factors through H = T or H = Id −T , i.e. Id H p (H q ) E / H p (H q ) O P H p (H q ) H kEkkP k ≤ C, / H p (H q ) where C = C(kT k). Capon invents a specific bi-parameterPGamlen-Gaudet selection process which leads to a block basis bI×J = K×L∈BI×J hK×L such that HbI×J = αI×J bI×J + small error, and the projection Qf = P hf,bI×J i I×J∈R kbI×J k22 bI×J αI×J ≥ 1/2. is bounded on H p (H q ). Capon’s theorem: H p (H q ) is primary (augmented) Theorem Let 1 < p, q < ∞ or p = q = 1. For any operator T the identity on H p (H q ) factors through H = T or H = Id −T , i.e. Id H p (H q ) E / H p (H q ) O P H p (H q ) H kEkkP k ≤ C, / H p (H q ) where C = C(kT k). Capon invents a specific bi-parameterPGamlen-Gaudet selection process which leads to a block basis bI×J = K×L∈BI×J hK×L such that HbI×J = αI×J bI×J + small error, and the projection Qf = P hf,bI×J i I×J∈R kbI×J k22 bI×J αI×J ≥ 1/2. is bounded on H p (H q ). Localization in H 1 (H 1 ) 1 (H 1 ) = span{h −N } ⊂ H 1 (H 1 ). HN I×J : |I|, |J| ≥ 2 N Theorem (R. L. & P. F. X. M.) For each n ∈ N there exists N (n) ∈ N, so that for each bounded linear 1 (H 1 ) → H 1 (H 1 ) we have operator T : HN N N N Hn1 (Hn1 ) E Id P 1 (H 1 ) HN N / H 1 (H 1 ) n O n H kEkkP k ≤ C, / H 1 (H 1 ) N N where H = T or H = Id −T , and C = C(kT k). • We need quantitative, finite dimensional estimates. • This leads to a combinatorial problem for coloured dyadic rectangles. Localization in H 1 (H 1 ) 1 (H 1 ) = span{h −N } ⊂ H 1 (H 1 ). HN I×J : |I|, |J| ≥ 2 N Theorem (R. L. & P. F. X. M.) For each n ∈ N there exists N (n) ∈ N, so that for each bounded linear 1 (H 1 ) → H 1 (H 1 ) we have operator T : HN N N N Hn1 (Hn1 ) E Id P 1 (H 1 ) HN N / H 1 (H 1 ) n O n H kEkkP k ≤ C, / H 1 (H 1 ) N N where H = T or H = Id −T , and C = C(kT k). • We need quantitative, finite dimensional estimates. • This leads to a combinatorial problem for coloured dyadic rectangles. Localization in H 1 (H 1 ) 1 (H 1 ) = span{h −N } ⊂ H 1 (H 1 ). HN I×J : |I|, |J| ≥ 2 N Theorem (R. L. & P. F. X. M.) For each n ∈ N there exists N (n) ∈ N, so that for each bounded linear 1 (H 1 ) → H 1 (H 1 ) we have operator T : HN N N N Hn1 (Hn1 ) E Id P 1 (H 1 ) HN N / H 1 (H 1 ) n O n H kEkkP k ≤ C, / H 1 (H 1 ) N N where H = T or H = Id −T , and C = C(kT k). • We need quantitative, finite dimensional estimates. • This leads to a combinatorial problem for coloured dyadic rectangles. Localization in H 1 (H 1 ) 1 (H 1 ) = span{h −N } ⊂ H 1 (H 1 ). HN I×J : |I|, |J| ≥ 2 N Theorem (R. L. & P. F. X. M.) For each n ∈ N there exists N (n) ∈ N, so that for each bounded linear 1 (H 1 ) → H 1 (H 1 ) we have operator T : HN N N N Hn1 (Hn1 ) E Id P 1 (H 1 ) HN N / H 1 (H 1 ) n O n H kEkkP k ≤ C, / H 1 (H 1 ) N N where H = T or H = Id −T , and C = C(kT k). • We need quantitative, finite dimensional estimates. • This leads to a combinatorial problem for coloured dyadic rectangles. Localization in H 1 (H 1 ) 1 (H 1 ) = span{h −N } ⊂ H 1 (H 1 ). HN I×J : |I|, |J| ≥ 2 N Theorem (R. L. & P. F. X. M.) For each n ∈ N there exists N (n) ∈ N, so that for each bounded linear 1 (H 1 ) → H 1 (H 1 ) we have operator T : HN N N N Hn1 (Hn1 ) E Id P 1 (H 1 ) HN N / H 1 (H 1 ) n O n H kEkkP k ≤ C, / H 1 (H 1 ) N N where H = T or H = Id −T , and C = C(kT k). • We need quantitative, finite dimensional estimates. • This leads to a combinatorial problem for coloured dyadic rectangles. We aim for a block basis bI×J = P K×L∈BI×L 1 (H 1 ) such that hK×L ∈ HN N • it well spans a copy of Hn1 (Hn1 ), • HbI×J = αI×J bI×J + small error, αI×J ≥ 1/2, p q • and Q is bounded on HN (HN ) with universal constants, where Qf = X hf, bI×J i bI×J . kbI×J k22 I×J∈R To get there we use a quantitative, finite dimensional replacement for hx, rm ⊗ h[0,1] i → 0 where {rm } is the Rademacher system. as m → ∞, We aim for a block basis bI×J = P K×L∈BI×L 1 (H 1 ) such that hK×L ∈ HN N • it well spans a copy of Hn1 (Hn1 ), • HbI×J = αI×J bI×J + small error, αI×J ≥ 1/2, p q • and Q is bounded on HN (HN ) with universal constants, where Qf = X hf, bI×J i bI×J . kbI×J k22 I×J∈R To get there we use a quantitative, finite dimensional replacement for hx, rm ⊗ h[0,1] i → 0 where {rm } is the Rademacher system. as m → ∞, We aim for a block basis bI×J = P K×L∈BI×L 1 (H 1 ) such that hK×L ∈ HN N • it well spans a copy of Hn1 (Hn1 ), • HbI×J = αI×J bI×J + small error, αI×J ≥ 1/2, p q • and Q is bounded on HN (HN ) with universal constants, where Qf = X hf, bI×J i bI×J . kbI×J k22 I×J∈R To get there we use a quantitative, finite dimensional replacement for hx, rm ⊗ h[0,1] i → 0 where {rm } is the Rademacher system. as m → ∞, We aim for a block basis bI×J = P K×L∈BI×L 1 (H 1 ) such that hK×L ∈ HN N • it well spans a copy of Hn1 (Hn1 ), • HbI×J = αI×J bI×J + small error, αI×J ≥ 1/2, p q • and Q is bounded on HN (HN ) with universal constants, where Qf = X hf, bI×J i bI×J . kbI×J k22 I×J∈R To get there we use a quantitative, finite dimensional replacement for hx, rm ⊗ h[0,1] i → 0 where {rm } is the Rademacher system. as m → ∞, We aim for a block basis bI×J = P K×L∈BI×L 1 (H 1 ) such that hK×L ∈ HN N • it well spans a copy of Hn1 (Hn1 ), • HbI×J = αI×J bI×J + small error, αI×J ≥ 1/2, p q • and Q is bounded on HN (HN ) with universal constants, where Qf = X hf, bI×J i bI×J . kbI×J k22 I×J∈R To get there we use a quantitative, finite dimensional replacement for hx, rm ⊗ h[0,1] i → 0 where {rm } is the Rademacher system. as m → ∞, Combinatorial covering lemma Let τ, δ > 0, x ∈ H 1 (H 1 ) and kxk ≤ 1. There exists a large collection of pairwise disjoint dyadic intervals K such that |hx, hK×[0,1] i| ≤ τ |K| and |K| ≥ δ 2 τ 2 (m ≥ − log2 δ 2 τ 2 ). Combinatorial covering lemma Let τ, δ > 0, x ∈ H 1 (H 1 ) and kxk ≤ 1. There exists a large collection of pairwise disjoint dyadic intervals K such that |hx, hK×[0,1] i| ≤ τ |K| and |K| ≥ δ 2 τ 2 (m ≥ − log2 δ 2 τ 2 ). Combinatorial covering lemma Let τ, δ > 0, x ∈ H 1 (H 1 ) and kxk ≤ 1. There exists a large collection of pairwise disjoint dyadic intervals K such that |hx, hK×[0,1] i| ≤ τ |K| and |K| ≥ δ 2 τ 2 [0, 1] [0, 1] P K |K| ≥ (1 − δ) (m ≥ − log2 δ 2 τ 2 ). Combinatorial covering lemma Let τ, δ > 0, x ∈ H 1 (H 1 ) and kxk ≤ 1. There exists a large collection of pairwise disjoint dyadic intervals K such that |hx, hK×[0,1] i| ≤ τ |K| and |K| ≥ δ 2 τ 2 [0, 1] (m ≥ − log2 δ 2 τ 2 ). [0, 1] [0, 1] [0, 1] P K |K| ≥ (1 − δ) P L |L| ≥ (1 − δ) Combinatorial covering lemma Let τ, δ > 0, x ∈ H 1 (H 1 )∗ and kxk ≤ 1. There exists a large collection of pairwise disjoint dyadic intervals K such that |hx, hK×[0,1] i| ≤ τ |K| and |K| ≥ δ 2 τ 2 [0, 1] (m ≥ − log2 δ 2 τ 2 ). [0, 1] [0, 1] [0, 1] P K |K| ≥ (1 − δ) P L |L| ≥ (1 − δ) Andrew 2D Theorem (R. L. & P. F. X. M.) Let 1 ≤ p, q < ∞, δ > 0 and T : H p (H q ) → H p (H q ) be a linear operator with large diagonal, i.e. hT hI×J , hI×J i ≥ δ|I × J|. Then we have H p (H q ) E Id P H p (H q ) / H p (H q ) O T kEkkP k ≤ C/δ, / H p (H q ) where the constant C > 0 is universal. By Capon’s bi-parameter construction (BI×J ) and a random choice of signs P (ε) εI×J , there exists a block basis bI×J = K∈BI×J εK×L hK×L of the Haar system such that (ε) (ε) T bI×J = αI×J bI×J + small error, αI×J ≥ δ. P hf,b i The orthogonal projection Qf = I∈D kb I×Jk2 bI×J is bounded on H p (H q ). I×J 2 Andrew 2D Theorem (R. L. & P. F. X. M.) Let 1 ≤ p, q < ∞, δ > 0 and T : H p (H q ) → H p (H q ) be a linear operator with large diagonal, i.e. hT hI×J , hI×J i ≥ δ|I × J|. Then we have H p (H q ) E Id P H p (H q ) / H p (H q ) O T kEkkP k ≤ C/δ, / H p (H q ) where the constant C > 0 is universal. By Capon’s bi-parameter construction (BI×J ) and a random choice of signs P (ε) εI×J , there exists a block basis bI×J = K∈BI×J εK×L hK×L of the Haar system such that (ε) (ε) T bI×J = αI×J bI×J + small error, αI×J ≥ δ. P hf,b i The orthogonal projection Qf = I∈D kb I×Jk2 bI×J is bounded on H p (H q ). I×J 2 Andrew 2D Theorem (R. L. & P. F. X. M.) Let 1 ≤ p, q < ∞, δ > 0 and T : H p (H q ) → H p (H q ) be a linear operator with large diagonal, i.e. hT hI×J , hI×J i ≥ δ|I × J|. Then we have H p (H q ) E Id P H p (H q ) / H p (H q ) O T kEkkP k ≤ C/δ, / H p (H q ) where the constant C > 0 is universal. By Capon’s bi-parameter construction (BI×J ) and a random choice of signs P (ε) εI×J , there exists a block basis bI×J = K∈BI×J εK×L hK×L of the Haar system such that (ε) (ε) T bI×J = αI×J bI×J + small error, αI×J ≥ δ. P hf,b i The orthogonal projection Qf = I∈D kb I×Jk2 bI×J is bounded on H p (H q ). I×J 2 Andrew 2D Theorem (R. L. & P. F. X. M.) Let 1 ≤ p, q < ∞, δ > 0 and T : H p (H q ) → H p (H q ) be a linear operator with large diagonal, i.e. hT hI×J , hI×J i ≥ δ|I × J|. Then we have H p (H q ) E Id P H p (H q ) / H p (H q ) O T kEkkP k ≤ C/δ, / H p (H q ) where the constant C > 0 is universal. By Capon’s bi-parameter construction (BI×J ) and a random choice of signs P (ε) εI×J , there exists a block basis bI×J = K∈BI×J εK×L hK×L of the Haar system such that (ε) (ε) T bI×J = αI×J bI×J + small error, αI×J ≥ δ. P hf,b i The orthogonal projection Qf = I∈D kb I×Jk2 bI×J is bounded on H p (H q ). I×J 2 Andrew 2D Theorem (R. L. & P. F. X. M.) Let 1 ≤ p, q < ∞, δ > 0 and T : H p (H q ) → H p (H q ) be a linear operator with large diagonal, i.e. hT hI×J , hI×J i ≥ δ|I × J|. Then we have H p (H q ) E Id P H p (H q ) / H p (H q ) O T kEkkP k ≤ C/δ, / H p (H q ) where the constant C > 0 is universal. By Capon’s bi-parameter construction (BI×J ) and a random choice of signs P (ε) εI×J , there exists a block basis bI×J = K∈BI×J εK×L hK×L of the Haar system such that (ε) (ε) T bI×J = αI×J bI×J + small error, αI×J ≥ δ. P hf,b i The orthogonal projection Qf = I∈D kb I×Jk2 bI×J is bounded on H p (H q ). I×J 2 Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q ) • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). • Full characterization in 1D for H p by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S I⊂E,|I|=2−i T bI exists almost everywhere, R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook ? • Full characterization in 2D for H p (H q )?? • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q )? • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ). Summary Let T : H p (H q ) → H p (H q ) be a linear operator. When does the identity factor through T (or Id −T )? • Results for H = T or H = Id −T : • Capon’s theorem for H p (H q ), 1 1 • local result in HN (HN ) (R. L. & P. F. X. M.). • Positive results, if T has a large diagonal: • Andrew 1D for H p , • Andrew 2D for H p (H q ) (R. L. & P. F. X. M.), p q • local version of Andrew 2D for HN (HN ) (R. L. & P. F. X. M.). p • Full characterization in 1D for H by property A (JMST): • {bI : I ∈ D} equivalent to the Haar system, 2 P • Λ(E) = limi→∞ S T b exists almost everywhere, −i I I⊂E,|I|=2 R1 • inf n 0 maxE∈D, |E|=2−n Λp/2 (E) dt > 0. Outlook • Full characterization in 2D for H p (H q )? • 2D version of property A, • Projection Q onto more general block basis bI×J in H p (H q ).