On Kato’s Square Root Problem Moritz Egert WIAS Berlin, February 11, 2015 T. Kato 1960s: Non-autonomous parabolic evolution equation d u(t)(x) + A(t)u(t)(x) = 0 (t > 0, x ∈ Ω), dt u(0) = u ∈ L2 (Ω). 0 I I A(t) ∼ − ∇x · µ(t, x)∇x via elliptic form a(t) : V(Ω) × V(Ω) → C. u(t)(x) = e−tA u0 (x) if A(t) = A for all t > 0. T. Kato 1960s: Non-autonomous parabolic evolution equation d u(t)(x) + A(t)u(t)(x) = 0 (t > 0, x ∈ Ω), dt u(0) = u ∈ L2 (Ω). 0 I I A(t) ∼ − ∇x · µ(t, x)∇x via elliptic form a(t) : V(Ω) × V(Ω) → C. u(t)(x) = e−tA u0 (x) if A(t) = A for all t > 0. T. Kato 1960s: Non-autonomous parabolic evolution equation d u(t)(x) + A(t)u(t)(x) = 0 (t > 0, x ∈ Ω), dt u(0) = u ∈ L2 (Ω). 0 I I A(t) ∼ − ∇x · µ(t, x)∇x via elliptic form a(t) : V(Ω) × V(Ω) → C. u(t)(x) = e−tA u0 (x) if A(t) = A for all t > 0. Kato Square Root Problem (1961) “We do not know whether or not D(A1/2 ) = D(A∗1/2 ). This is perhaps not true in general. But the question is open even when A is regularly accretive. In this case it appears reasonable to suppose that both D(A1/2 ) and D(A∗1/2 ) coincide with D(a), where a is the regular sesquilinear form which defines A.” T. Kato 1960s: Non-autonomous parabolic evolution equation d u(t)(x) + A(t)u(t)(x) = 0 (t > 0, x ∈ Ω), dt u(0) = u ∈ L2 (Ω). 0 I I A(t) ∼ − ∇x · µ(t, x)∇x via elliptic form a(t) : V(Ω) × V(Ω) → C. u(t)(x) = e−tA u0 (x) if A(t) = A for all t > 0. Kato Square Root Problem (1961) “We do not know whether or not D(A1/2 ) = D(A∗1/2 ). This is perhaps not true in general. But the question is open even when A is regularly accretive. In this case it appears reasonable to suppose that both D(A1/2 ) and D(A∗1/2 ) coincide with D(a), where a is the regular sesquilinear form which defines A.” I I Counterexamples by Lions 1962, Mc Intosh 1982 Specialize to divergence-form operators. Setup Let I Ω ⊆ Rd domain, D ⊆ ∂ Ω closed, µ ∈ L∞ (Ω) A ∼ −∇ · µ∇ accretive operator on L2 (Ω) associated with Z 1,2 W1,2 (Ω) × W (u, v ) 7→ µ∇u · ∇v . D D (Ω) → C, I Ω I A1/2 square root of A defined by e.g. Z ∞ 1 A1/2 u = t −1/2 A(t + A)−1 dt. π 0 Kato conjecture It holds D(A1/2 ) = W1,2 D (Ω) with equivalent norms. D Why do we care about the Kato conjecture? Philosophy I Elliptic non-regularity results D(A) * W2,2 (Ω). I Kato Conjecture ∼ optimal Sobolev regularity for A1/2 . Why do we care about the Kato conjecture? Philosophy I Elliptic non-regularity results D(A) * W2,2 (Ω). I Kato Conjecture ∼ optimal Sobolev regularity for A1/2 . Ex. 1: Elliptic equations on Rd+ ∂2 d u(t)(x) + ∇ · µ(x)∇u(t, x) = 0 (t > 0, x ∈ R ), 2 ∂t u(0, x) = u0 (x) ∈ W1,2 (Rd ). 1/2 I Solution u(t, x) = e−tA I Kato conjecture ∼ Rellich inequality “k∂t u|t=0 k2 ∼ k∇u0 k2 ”. u0 (x). Why do we care about the Kato conjecture? Philosophy I Elliptic non-regularity results D(A) * W2,2 (Ω). I Kato Conjecture ∼ optimal Sobolev regularity for A1/2 . Ex. 1: Elliptic equations on Rd+ ∂2 d u(t)(x) + ∇ · µ(x)∇u(t, x) = 0 (t > 0, x ∈ R ), 2 ∂t u(0, x) = u0 (x) ∈ W1,2 (Rd ). 1/2 I Solution u(t, x) = e−tA I Kato conjecture ∼ Rellich inequality “k∂t u|t=0 k2 ∼ k∇u0 k2 ”. u0 (x). Ex. 2: Maximal parabolic regularity (e.g. Haller-Dintelmann-Rehberg) I In Lp -setting study parabolic equation ( I d u(t) + Au(t) = f dt u(0) = 0. (0 < t < T ), Goal: Transport Max. Reg. from Lp (Ω) to W−1,p (Ω). D Ex. 2: Maximal parabolic regularity (e.g. Haller-Dintelmann-Rehberg) I In Lp -setting study parabolic equation ( I d u(t) + Au(t) = f dt u(0) = 0. (0 < t < T ), Goal: Transport Max. Reg. from Lp (Ω) to W−1,p (Ω). D 0 I Lp -Kato conjecture ∼ (−∇ · µ> ∇)1/2 : 1,p0 WD (Ω) 0 → Lp (Ω) isom. Ex. 2: Maximal parabolic regularity (e.g. Haller-Dintelmann-Rehberg) I In Lp -setting study parabolic equation ( I (0 < t < T ), Goal: Transport Max. Reg. from Lp (Ω) to W−1,p (Ω). D 0 I Lp -Kato I d u(t) + Au(t) = f dt u(0) = 0. conjecture ∼ (−∇ · µ> ∇)1/2 : 1,p0 WD (Ω) 0 → Lp (Ω) isom. Adjoint (−∇ · µ∇)1/2 : Lp (Ω) → WD−1,p (Ω) isomorphism that commutes with parabolic solution operator d −1 +A . dt Ex. 2: Maximal parabolic regularity (e.g. Haller-Dintelmann-Rehberg) I In Lp -setting study parabolic equation ( I (0 < t < T ), Goal: Transport Max. Reg. from Lp (Ω) to W−1,p (Ω). D 0 I Lp -Kato I d u(t) + Au(t) = f dt u(0) = 0. conjecture ∼ (−∇ · µ> ∇)1/2 : 1,p0 WD (Ω) 0 → Lp (Ω) isom. Adjoint (−∇ · µ∇)1/2 : Lp (Ω) → WD−1,p (Ω) isomorphism that commutes with parabolic solution operator d −1 +A . dt Many further examples, e.g. Cauchy-Integral along Lipschitz curve, hyperbolic wave equations, . . . . Positive answers Self-adjoint operators Positive answers Self-adjoint operators X Positive answers Self-adjoint operators X Whole space Ω = Rd I I d = 1: Coifman - Mc Intosh - Meyer ’82. d ≥ 2: Auscher-Hofmann-Lacey-Mc Intosh-Tchamitchian ’01, Axelsson-Keith-Mc Intosh ’06. Positive answers Self-adjoint operators Whole space Ω = Rd I I X X d = 1: Coifman - Mc Intosh - Meyer ’82. d ≥ 2: Auscher-Hofmann-Lacey-Mc Intosh-Tchamitchian ’01, Axelsson-Keith-Mc Intosh ’06. Positive answers Self-adjoint operators Whole space Ω = Rd I I X X d = 1: Coifman - Mc Intosh - Meyer ’82. d ≥ 2: Auscher-Hofmann-Lacey-Mc Intosh-Tchamitchian ’01, Axelsson-Keith-Mc Intosh ’06. Bounded domains I Ω Lipschitz, D ∈ {∅, ∂ Ω}: Auscher-Tchamitchian ’03, ’01 (p 6= 2). I Ω smooth, smooth D ! ∂ Ω \ D interface: Axelsson-Keith-Mc Intosh ’06. I Ω Lipschitz around ∂ Ω \ D: Auscher-Badr-Haller-Dintelmann-Rehberg ’12 (p 6= 2). Kato for mixed boundary conditions Theorem (E.-Haller-Dintelmann-Tolksdorf ’14) Suppose I Ω ⊆ Rd bounded d-Ahlfors regular domain, I D ⊆ ∂ Ω closed and (d − 1)-Ahlfors regular, I Ω Lipschitz around ∂ Ω \ D. Then D(A1/2 ) = W1,2 D (Ω) with kA1/2 uk2 ∼ k∇uk2 . I First formulated by J.-L. Lions 1962. I For rough (= L∞ ) coefficients new even on Lipschitz domains. Some ideas of the proof 1 First-order approach via perturbed Dirac operators á la AKM ’06, H∞ -functional calculus. Some ideas of the proof 1 First-order approach via perturbed Dirac operators á la AKM ’06, H∞ -functional calculus. 2 Getting rid of the coefficients via Reduction-Theorem (E.-Haller-Dintelmann-Tolksdorf ’14) s In essence, the following holds: If D (−∆V ) ,→ H2s,2 (Ω) for some s > 12 , then D(A1/2 ) = W1,2 D (Ω). Some ideas of the proof 1 First-order approach via perturbed Dirac operators á la AKM ’06, H∞ -functional calculus. 2 Getting rid of the coefficients via Reduction-Theorem (E.-Haller-Dintelmann-Tolksdorf ’14) s In essence, the following holds: If D (−∆V ) ,→ H2s,2 (Ω) for some s > 12 , then D(A1/2 ) = W1,2 D (Ω). Extrapolate Kato for −∆V =⇒ Kato property for general A Some ideas of the proof 1 First-order approach via perturbed Dirac operators á la AKM ’06, H∞ -functional calculus. 2 Getting rid of the coefficients via Reduction-Theorem (E.-Haller-Dintelmann-Tolksdorf ’14) s In essence, the following holds: If D (−∆V ) ,→ H2s,2 (Ω) for some s > 12 , then D(A1/2 ) = W1,2 D (Ω). Extrapolate Kato for −∆V =⇒ Kato property for general A geometry, potential theory ! harmonic analysis 3 D((−∆V )1/2 ) = W1,2 D (Ω) by self-adjointness. Extrapolate by Sneiberg’s stability theorem and the following result. Theorem (E.-Haller-Dintelmann-Tolksdorf ’14) Let θ ∈ (0, 1) and s0 , s1 ∈ ( 12 , 32 ). Put sθ := (1 − θ)s0 + θs1 . Then, 1,2 W1,2 (Ω) = H D D (Ω) s0 ,2 s1 ,2 HD (Ω), HD (Ω) θ = HsDθ ,2 (Ω). ( 2 1,2 L (Ω), HD (Ω) θ = 1 Hθ,2 (Ω), if θ > D 2, Hθ,2 (Ω), if θ < 12 . 3 D((−∆V )1/2 ) = W1,2 D (Ω) by self-adjointness. Extrapolate by Sneiberg’s stability theorem and the following result. Theorem (E.-Haller-Dintelmann-Tolksdorf ’14) Let θ ∈ (0, 1) and s0 , s1 ∈ ( 12 , 32 ). Put sθ := (1 − θ)s0 + θs1 . Then, 1,2 W1,2 (Ω) = H D D (Ω) s0 ,2 s1 ,2 HD (Ω), HD (Ω) θ = HsDθ ,2 (Ω). ( 2 1,2 L (Ω), HD (Ω) θ = 4 1 Hθ,2 (Ω), if θ > D 2, Hθ,2 (Ω), if θ < 12 . 1 In fact, D((−∆V )s ) = H2s,2 (Ω) for | D 2 − s| < ε. Elliptic BVPs on cylindrical domains Elliptic mixed BVP −divt,x µ(x)∇t,x U = 0 (R+ × Ω) U=0 (R+ × D) ∂νµ U = 0 (R+ × N) ∂νµ U = f ∈ L2 ({0} × Ω) Elliptic BVPs on cylindrical domains Elliptic mixed BVP −divt,x µ(x)∇t,x U = 0 (R+ × Ω) U=0 (R+ × D) ∂νµ U = 0 (R+ × N) ∂νµ U = f ∈ L2 l ∂νµ U F ∼ ∇x U ({0} × Ω) First order equation ∗ 0 (−∇V ) ∂t F + BF = 0 −∇V 0 | {z } D F (0)⊥ = f (t > 0) Elliptic BVPs on cylindrical domains Elliptic mixed BVP −divt,x µ(x)∇t,x U = 0 (R+ × Ω) U=0 (R+ × D) ∂νµ U = 0 (R+ × N) ∂νµ U = f ∈ L2 l ∂νµ U F ∼ ∇x U ({0} × Ω) First order equation ∗ 0 (−∇V ) ∂t F + BF = 0 −∇V 0 | {z } (t > 0) D F (0)⊥ = f L2loc (R+ ; L2 (Ω)) setting Semigroup solutions via DB-formalism DB has bounded H∞ -calculus on H = R(DB) (Kato Technology). Theorem (Auscher-E. ’14) 1 For every F (0) ∈ H+ := 1C+ (DB)H a solution to the first-order system is F (t) = e−tDB F (0) (t ≥ 0). ∂ U Via F ∼ νµ these functions are in one-to-one correspon∇x U dence with weak solutions U such that f∗ (|∇t,x U|) ∈ L2 (R+ × Ω). N 2 If µ is either block-diagonal or Hermitean, then for each f ∈ L2 (Ω) there exists a unique such solution u. Thank you for your attention! Kato Square Root Problem Literature [1] T. K ATO. Fractional powers of dissipative operators. J. Math. Soc. Japan 13 (1961), 246–274. [2] P. AUSCHER, S. H OFMANN, M. L ACEY, A. M C I NTOSH, and P. T CHAMITCHIAN. The solution of the Kato square root problem for second order elliptic operators on Rn . Ann. of Math. (2) 156 (2002), no. 2, 633–654. [3] A. A XELSSON, S. K EITH, and A. M C I NTOSH. Quadratic estimates and functional calculi of perturbed Dirac operators. Invent. Math. 163 (2006), no. 3, 455–497. [4] M. E GERT, R. H ALLER -D INTELMANN, and P. TOLKSDORF. The Kato Square Root Problem follows from an extrapolation property of the Laplacian. Submitted. [5] M. E GERT, R. H ALLER -D INTELMANN, and P. TOLKSDORF. The Kato Square Root Problem for mixed boundary conditions. J. Funct. Anal. 267 (2014), no. 5, 1419–1461. [6] P. AUSCHER, A. A XELSSON, and A. M C I NTOSH. Solvability of elliptic systems with square integrable boundary data. Ark. Mat. 48 (2010), no. 2, 253–287.