Directional oscillations and concentrations and weak↔strong compactness via microlocal compactness forms Filip Rindler Cambridge Centre for Analysis (CCA) University of Cambridge (DAMTP) F.Rindler@maths.cam.ac.uk 6 December 2012 Oscillations & concentrations Prototypical oscillation: uj (x) = sin(jx · n) , , n , ··· 0 In Fourier space: mass wanders out to ∞: Prototypical concentration in Lp : uj (x) = j d/p h(jx) , , , ··· Why study oscillations / concentrations? Weak↔strong compactness: p (uj ) ⊂ L bounded, Vitali’s Convergence Thm uj → u in measure ⇔ no oscillations, =⇒ uj → u (uj ) p-equiintegrable ⇔ no concentrations Compensated compactness (Tartar’s framework): uj * u in Lp , d X ∂u ??? j (k) −1,p Auj := A = 0 in W , =⇒ uj → u ∂x k k=1 uj (x) ∈ Z (x) a.e. Efficient description of microstructure (e.g. for minimizing sequences) Physical meaning of weak/strong convergence: Assume: (uj ) ⊂ L2 sequence of “measurements”. Weak convergence uj * u: Z ϕ · (uj − u) dx Parseval Z ϕ̂ · F [uj − u] dξ, = Weak convergence is band-limited Strong convergence uj → u: Z |uj − u|2 dx ∀ϕ ∈ S. Measure with “finite precision”. Parseval = Strong convergence is not band-limited: Z |F[uj − u]|2 dξ. Measure with “infinite precision”. Conclusion: Different types of convergence have different physical meanings. The study of compactness is related to the “physical” properties of oscillations & concentrations. Tools to systematically study oscillations & concentrations Measure compactness / oscillations / concentrations for a weakly compact generating sequence (uj ) ⊂ Lp : Location: Scalar defect measures w*-limj→∞ |uj |p Ld Location & value-distribution (oscillations): Young measures (Young ’37,’42) Location & value-distribution (oscillations & concentrations): Generalized Young (DiPerna–Majda) measures (DiPerna & Majda ’87) Location & direction: H-measures (Tartar ’90, Gérard ’91) Location & value-distribution & direction: Microlocal compactness forms (MCFs) ←− in this work (2012). Features of MCFs Represent limits of functionals (essentially contains generalized Young measure). Preserve directional information of oscillations/concentrations preserved (contains H-measure). Easy to “read off” pointwise constraints (u(x) ∈ Z (x)) and differential constraints (Auj = 0) on generating sequence. Allow an analogue of the wavefront set from microlocal analysis, but with respect to weak↔strong convergence (not C∞ -regularity). Hierarchy of microstructure (e.g. laminates) reflected in MCF. Allow relaxation of anisotropic functionals. Can be defined for all Lp -spaces (H-measure only for L2 ). Have the same compactness as weak convergence. Definition of MCFs For h ∈ C(Ω × CN ; CN ): (S p−1 h)(x, w ) := (1 − |w |)p−1 h x, w , 1 − |w | (x, w ) ∈ Ω × BN . Space Fp of test functions: Fp (Ω; CN ) := f ∈ C(Ω × CN × CN ; C) : f (x, z, q) = h(x, z) · q and S p−1 h ∈ C(Ω × BN ; CN ) . Fourier multiplier with symbol Ψ ∈ M := C∞ (Sd−1 ; CN×N ): ξ û(ξ) Tψ : Lp (Ω; CN ) → Lp (Ω; CN ), Tψ [u] := F −1 Ψ |ξ| d Cut-off function η ∈ C∞ c (R ) with η ≡ 1 on B(0, 1), supp η ⊂ B(0, 2). d ηR (ξ) := η(ξ/R) for ξ ∈ R and any R > 1. Microlocal compactness form ω ∈ MCFp (Ω; CN ) is a sesquilinear form Z f ⊗ Ψ, ω = ω(f , Ψ) = lim lim h( q, uj ) · T(1−ηR )Ψ [uj ] dx R→∞ j→∞ p N Ω for all f ∈ F (Ω; C ), Ψ ∈ M. (Well-defined?) Existence Theorem Theorem (R. 2012) Let (uj ) ⊂ Lp (Ω; CN ) be norm-bounded. Then, after selecting a subsequence, there exist an MCF ω ∈ MCFp (Ω; CN ) such that Z f ⊗ Ψ, ω = lim lim h( q, uj ) · T(1−ηR )Ψ [uj ] dx R→∞ j→∞ p Ω N for all f ∈ F (Ω; C ) with f (x, z, q) = h(x, z) · q and Ψ ∈ M. Moreover: (A) Independence. The above limit (and hence ω) is independent of the choice of η and the sequence R → ∞. (B) Basic estimate. For all f , Ψ as above it holds that f ⊗ Ψ, ω ≤ C · supj k1 + |uj |kpp · kf kFp · kΨkCd , where C = Cd max{p, (p − 1)−1 } with a dimensional constant Cd . Parts of the main definition formula h( q, uj ) ∈ Lp/(p−1) , T(1−ηR )Ψ [uj ] ∈ Lp . Product in L1 2 concentrations expressed. In L , Parseval yields: Z Z ξ ûj (ξ) dξ. h( q, uj )·T(1−ηR )Ψ [uj ] dx = F h( q, uj ) (ξ) · (1 − ηR (ξ))Ψ |ξ| Ω Ω Representation Theorem Theorem (R. 2012) The MCF ω ∈ MCFp (Ω; CN ) from the preceding Existence Theorem can be considered as a triple ω = (ωx , λω , ωx∞ ) consisting of (i) a parametrized family (ωx )x∈Ω of continuous (complex-valued) sesquilinear forms ωx : C(CN ; CN ) × M → C, (ii) a positive and finite measure on Ω, and (iii) a parametrized family (ωx∞ )x∈Ω of continuous (complex-valued) sesquilinear forms ωx∞ : C(∂BN ; CN ) × M → C. Then it holds that Z Z ∞ f ⊗ Ψ, ω = h(x, q) ⊗ Ψ, ωx dx + h (x, q) ⊗ Ψ, ωx∞ dλω (x). {z } {z } Ω| Ω| q q =ωx∞ (h∞ (x, ),Ψ) =ωx (h(x, ),Ψ) “oscillations” “concentrations” Recession function: h∞ (x, z) := limx 0 →x z 0 →z t→∞ h(x 0 , tz 0 ) (x ∈ Ω, z ∈ CN ). t p−1 Oscillations Consider: uj (x) = A1(0,θ) (jx · n0 − bjx · n0 c) + B 1(θ,1) (jx · n0 − bjx · n0 c), x ∈ Ω, where A, B ∈ CN , n0 ∈ Sd−1 and 1(0,θ) is the indicator function of (0, θ). What MCF does (uj ) generate? Lemma (Oscillation Lemma) Let w ∈ L∞ (R; CN ) be 1-periodic (the “profile function”) and assume that wj (x) := w (js) generates the homogeneous Young measure ν ∈ M1 (CN ). Then, the simple oscillation in direction n0 ∈ Sd−1 , uj (x) := w (jx · n0 ), x ∈ Ω, generates an microlocal compactness form ω = Ld Ω ⊗ (z − Z0 ) ν(dz) ⊗ δ ±n0 ∈ MCF2 (Ω; CN ), R1 where Z0 := −0 w ds is the average of w over one period cell and δ ±n0 := δ −n0 + δ +n0 . That is, for all f ∈ Fp (Ω; CN ), Ψ ∈ M, Z Z f ⊗ Ψ, ω = h(x, z) · Ψ(+n0 ) + Ψ(−n0 ) (z − Z0 ) dν(z) dx. Ω Oscillations II Example: Simple oscillation uj (x) = A1(0,θ) (bjx · n0 c) + B 1(θ,1) (bjx · n0 c), where A, B ∈ CN , n0 ∈ Sd−1 and x ∈ Ω, 1(0,θ) is the indicator function of (0, θ). Young measure: ν = θδA + (1 − θ)δB . MCF: By the Oscillation Lemma the sequence (uj ) generates ω = Ld Ω ⊗ θ(A − M)δA + (1 − θ)(B − M)δB ⊗ δ ±n0 ∈ MCF2 (Ω; CN ), where M := θA + (1 − θ)B. Proof of Oscillation Lemma: n0 Oscillations III Concentrations Let w ∈ Lp (Rd ) have compact support and w ≥ 0. Further, let Z0 ∈ CN with |Z0 | = 1. Define uj (x) := j d/p Z0 w (jx), x ∈ Rd . Then, uj generates the MCF ω = δ0 ⊗ Z0 δ∞Z0 ⊗ µ̄ ∈ MCFp (Rd ; CN ), where µ̄ is the surface measure (acting on Ψ) Z ∞ µ̄ = F |w |p−1 (tη) ŵ (tη) t d−1 dt Hd−1 Sd−1 (dη). 0 The above is a shorthand notation for the MCF ω acting on f (x, z, q) = h(x, z) · q ∈ Fp (Ω; CN ), Ψ ∈ M, as Z Ψ(ξ)Z0 dµ̄(ξ). f ⊗ Ψ = h∞ (0, Z0 ) · Sd−1 Remarks In the Existence Theorem, we can choose for example (↔ scalar defect measure) λω = w*-lim |uj |p Ld . j→∞ (?) The families (ωx )x , (ωx∞ )x are uniquely determined if we use the canonical choice (?). Proofs of existence/representation theorems use techniques from harmonic analysis, measure theory (disintegration/slicing of measures), and the theory of generalized Young measures (we use an argument similar to Kristensen & R. 2010, ARMA). The representation as before allows us to (in some sense) localize in both the x- and ξ-variable and so “circumvent the uncertainty principle”! Reason: We only look at infinite frequencies. Some properties of MCF Lemma Let (uj ) ⊂ Lp (Ω; CN ) generate the MCF ω ∈ MCFp (Ω; CN ) and uj * u in Lp (Ω; CN ). Then uj → u strongly if and only if ω = 0 The MCF ω measures the difference between weak and strong convergence. Young measures: Proposition Let (uj ) ⊂ Lp (Ω; CN ) generate the MCF ω ∈ MCFp (Ω; CN ) and the (generalized) Young measure ν ∈ Y(Ω; CN ). Assume further that uj * u = [ν]. Then, the knowledge of ω and of the limit u completely determine ν. H-measures: They are trivially contained (in the case uj * 0 in L2 ). Differential constraints Differential operator A: A := d X ∂ ∂xk A(k) k=1 Symbol: A(ξ) := Pd k=1 A(k) ξk . (Au = 0 ⇔ (2πi)A(ξ)û(ξ) = 0) Homogeneous symbol: A0 (ξ) := Pd k=1 A(k) ξk /|ξ|. Murat’s constant-rank property: rank ker A(ξ) = const for all ξ ∈ Sd−1 . Theorem (R. 2012) Let (uj ) ⊂ L2 (Rd ; CN ) with uj * u in L2 (Rd ; CN ) generate the MCF ω ∈ MCF2 (Rd ; CN ). Then, Auj → 0 in W−1,2 (Rd ; Cl ). if and only if 2 N f ⊗ ΨA0 , ω for all f ∈ F (Ω; C ) and all Ψ ∈ C bd/2c+1 =0 d−1 (S ; CN×l ). Higher-order laminates Homogeneous MCFs: ω ∈ MCFp (Ω; CN ) is called homogeneous if ωx , ωx∞ are constant in x and λω = αLd Ω for a constant α > 0. Proposition (Laminations) Let n0 ∈ Sd−1 . Assume that (i) uj * A = const and vj * B = const in Lp (Rd ; CN ), (ii) (uj ), (vj ) are p-equiintegrable, (iii) (uj ), (vj ) generate the homogeneous MCFs ω1 , ω2 ∈ MCFphom (CN ) and the homogeneous Young measures ν1 , ν2 ∈ M1 (CN ), respectively. Then, for any θ ∈ (0, 1) there exists a homogeneous microlocal compactness form ω̄ ∈ MCFphom (CN ) with ω̄ = θω1 + (1 − θ)ω2 + Ld ⊗ θ(A − X )ν1 + (1 − θ)(B − X )ν2 ⊗ δ ±n0 , | {z } | {z } faster scales slower scale where X := θA + (1 − θ)B. Proof: Similar argument as in Oscillation Lemma, and averaging. Hierarchy of scales reflected. Lamination example C n1 A B A n2 B A B A B A B A 1 − θ1 M A 1 − θ2 X θ2 θ1 B C C A B A B A B A M = θ1 A + (1 − θ1 )B X = θ2 M + (1 − θ2 )C Generated MCF: ω = θ2 ωA/B + (1 − θ2 )ωC + Ld ⊗ θ2 (M − X )ν1 + (1 − θ2 )(C − X )ν2 ⊗ δ ±n2 n = Ld Ω ⊗ θ2 θ1 (A − M)δA + θ2 (1 − θ1 )(B − M)δB ⊗ δ ±n1 + θ2 (M − X )(θ1 δA + (1 − θ1 )δB ) ⊗ δ ±n2 o + (1 − θ2 )(C − X )δC ⊗ δ ±n2 . Wavefront set Sphere compactification: σCN := CN ] ∞SN−1 . Wavefront set for ω ∈ MCF(Ω; CN ): WF(ω) ⊂ Ω × σCN × Sd−1 is the smallest closed subset A of Ω × σCN × Sd−1 with the property that for any f (x, z, q) = ϕ(x)g (z) · q ∈ Fp (Ω; CN ) and any multiplier Ψ ∈ M, supp ϕ ⊗ g ⊗ Ψ ⊂ Ac implies f ⊗ Ψ, ω = 0. Examples: Oscillations: ω = Ld Ω ⊗ θ(A − M)δA + (1 − θ)(B − M)δB ⊗ δ ±n0 : WF(ω) = Ω × {A, B} × {+n0 , −n0 }. Concentrations: ω = δ0 ⊗ Z0 δ∞Z0 ⊗ µ̄: WF(ω) = (0, ∞Z0 ) × supp µ̄, “Concentrations lie in ∞SN−1 ”. Compensated compactness Consider general framework (Tartar): uj * u in Lp , d X ∂uj Auj := A(k) = 0 in W−1,p , ∂x k k=1 uj (x) ∈ Z a.e., Z ⊂ CN closed. Symbol: A(ξ) := Pd k=1 A(k) ξk . (CC) (Au = 0 ⇔ (2πi)A(ξ)û(ξ) = 0) Theorem (R. 2012) Let (uj ) ⊂ Lp (Rd ; CN ) generate ω ∈ MCFp (Rd ; CN ) and let A be a linear PDE operator satisfying Murat’s constant-rank property, i.e. rank ker A(ξ) = const for all ξ ∈ Sd−1 . Moreover, assume (CC). Then, WF(ω) ⊂ Ξ := (x, z, ξ) ∈ Rd × σCN × Sd−1 : spanC Z ∩ ker A(ξ) 6= {0} . In particular, if Ξ = ∅, then ω = 0 and (uj ) is strongly compact. Compensated compactness example I: Gradients Let (uj ) ⊂ Lp (Ω; Rm×d ) generate the MCF ω ∈ MCFp (Ω; Cm×d ) and ( curl uj = 0 in W−1,p , uj (x) ∈ span{M} a.e., M ∈ Rm×d a fixed matrix. Theorem ⇒ WF(ω) ⊂ (x, z, ξ) ∈ Ω×σRm×d ×Sd−1 : spanC {M}∩ker A(ξ) 6= {0} , where ker A(ξ) = { a ⊗ ξ : a ∈ Cm }. (i) If rank M ≥ 2, then ω = 0 and (uj ) is strongly compact. (ii) If M = a ⊗ n0 for a ∈ Rm , n0 ∈ Sd−1 , then WF(ω) ⊂ (x, z, ξ) ∈ Ω × σRm×d × Sd−1 : ξ = ±n0 , Oscillations / concentrations have “direction ±n0 ”: Compensated compactness example II: Linear elasticity There exists a (2nd order) A acting on Rd×d sym -valued vector fields with 1 ∇v + ∇v T if and only if Au = 0, u= 2 p d m×d Let (uj ) ⊂ Lp (Ω; Rd×d sym ) generate the MCF ω ∈ MCF (R ; Csym ) with ( Auj = 0 in W−1,p , uj (x) ∈ span{M} a.e., M ∈ Rd×d sym a fixed matrix. Theorem ⇒ d−1 (x, z, ξ) ∈ Ω × σRd×d : spanC {M} ∩ ker A(ξ) 6= {0} , sym × S where ker A(ξ) = a ξ = 21 a ⊗ ξ + ξ ⊗ a : a ∈ Cd . WF(ω) ⊂ (i) If M ∈ / {a b}, then ω = 0 and (uj ) is strongly compact. (ii) If M = γ(a b) for some a, b ∈ Rd with |a| = |b| = 1, γ 6= 0, then d−1 WF(ω) ⊂ (x, z, ξ) ∈ Ω × σRd×d : ξ = ±a or ξ = ±b . sym × S Compensated compactness example II: Linear elasticity (continued) The case M = a b (continued): d−1 WF(ω) ⊂ (x, z, ξ) ∈ Ω × σRd×d : ξ = ±a or ξ = ±b . sym × S Oscillations and concentrations have direction ±a or ±b: = + Weak* lower semicontinuity of integral functionals depending on the symmetric gradient and in the space BD of functions of bounded deformation (R. 2011, ARMA) Summary & Outlook The theory of MCFs . . . . . . measures the difference between weak and strong compactness very weak regularity. . . . can be considered microlocal analysis for weak↔strong compactness much more adapted to nonlinear PDEs than classical microlocal analysis, which measures C∞ -regularity. . . . allows localizing singularities in x, ξ, and the target space (“circumvent the uncertainty principle”). . . . represents differential/pointwise constraints. . . . enables “geometric” proofs of compensated compactness. . . . reflects the hierarchy of microstructure. Outlook: Relaxation of integral functionals with anisotropy. Propagation of regularity / singularities for hyperbolic conservation laws. Finer investigations into shape of microstructure. Thank you for your attention!