Regularity for functionals involving perimeter

Transcription

1 Regularity for functionals involving perimeter Guido De Philippis, Jimmy Lamboley, Michel Pierre, Bozhidar Velichkov Université Paris Dauphine, CEREMADE 22/11/16, CIRM Shape optimization, Isoperimetric and Functional Inequalities Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 1 / 15

2 Statement of the problem We consider the optimization problem { } min P(Ω) + G(Ω) : Ω D, Ω = m, Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 2 / 15

3 Statement of the problem We consider the optimization problem { } min P(Ω) + G(Ω) : Ω D, Ω = m, where D is a (possibly unbounded) set in R d, m ]0, D [, Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 2 / 15

4 Statement of the problem We consider the optimization problem { } min P(Ω) + G(Ω) : Ω D, Ω = m, where D is a (possibly unbounded) set in R d, m ]0, D [, P is the perimeter in the sense of De Giorgi, Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 2 / 15

5 Statement of the problem We consider the optimization problem { } min P(Ω) + G(Ω) : Ω D, Ω = m, where D is a (possibly unbounded) set in R d, m ]0, D [, P is the perimeter in the sense of De Giorgi, { } P(Ω) := sup divφ dx : φ Cc 1 (R d ; R d ), φ 1 on R d = 1 Ω TV, Ω Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 2 / 15

6 Statement of the problem We consider the optimization problem { } min P(Ω) + G(Ω) : Ω D, Ω = m, where D is a (possibly unbounded) set in R d, m ]0, D [, P is the perimeter in the sense of De Giorgi, { } P(Ω) := sup divφ dx : φ Cc 1 (R d ; R d ), φ 1 on R d = 1 Ω TV, Ω G is one of the following: the Dirichlet energy E f, with respect to a (possibly sign-changing) function f L p (D); Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 2 / 15

7 Statement of the problem We consider the optimization problem { } min P(Ω) + G(Ω) : Ω D, Ω = m, where D is a (possibly unbounded) set in R d, m ]0, D [, P is the perimeter in the sense of De Giorgi, { } P(Ω) := sup divφ dx : φ Cc 1 (R d ; R d ), φ 1 on R d = 1 Ω TV, Ω G is one of the following: the Dirichlet energy E f, with respect to a (possibly sign-changing) function f L p (D); { } 1 E f (Ω) = min u 2 dx uf dx, u H0 1 (Ω) 2 R d R d Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 2 / 15

8 Statement of the problem We consider the optimization problem { } min P(Ω) + G(Ω) : Ω D, Ω = m, where D is a (possibly unbounded) set in R d, m ]0, D [, P is the perimeter in the sense of De Giorgi, { } P(Ω) := sup divφ dx : φ Cc 1 (R d ; R d ), φ 1 on R d = 1 Ω TV, Ω G is one of the following: the Dirichlet energy E f, with respect to a (possibly sign-changing) function f L p (D); { } 1 E f (Ω) = min u 2 dx uf dx, u H0 1 (Ω) 2 R d R d a spectral functional of the form F (λ 1(Ω),..., λ k (Ω)), where (λ 1(Ω),..., λ k (Ω)) are the first k eigenvalues of the Dirichlet-Laplacian and F : R k R is locally Lipschitz continuous and increasing in each variable. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 2 / 15

9 Examples If G = 0, D = R d. Isoperimetric inequality. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 3 / 15

10 Examples If G = 0, D = R d. Isoperimetric inequality. The ball is solution. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 3 / 15

11 Examples If G = 0, D = R d. Isoperimetric inequality. The ball is solution. If G = E 1 or λ 1. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 3 / 15

12 Examples If G = 0, D = R d. Isoperimetric inequality. The ball is solution. If G = E 1 or λ 1. Isoperimetric + Saint-Venant/Faber-Krahn inequalities. The ball is solution. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 3 / 15

13 Examples Numerical results by Bogosel-Oudet 2014 min { P(Ω) + λ k (Ω), Ω R 2} no volume constraint Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 4 / 15

14 Examples Numerical results by Bogosel-Oudet 2014 min { SPECTRAL ANALYSIS - PERIMETER P(Ω) + λ k (Ω), Ω R 2} CONSTRAINT 15 no volume constraint λ 1 = λ 2 = λ 3 = (double) λ 4 = (double) λ 5 = (double) λ 6 = λ 7 = (double) λ 8 = (double) λ 9 = λ 10 = (double) λ 11 = (double) λ 12 = (triple) λ 13 = λ 14 = (double) λ 15 = Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 4 / 15

15 (double) (double) (double) Examples Numerical results by Bogosel-Oudet 2014 λ 11 = λ 12 = (double) (triple) λ 13 = λ 14 = (double) min { P(Ω) + λ k (Ω), Ω R 3}. Figure 1. Numerical optimizers for problem (1.3) in 2D λ 15 = λ 2(Ω) Per(Ω) = λ 3(Ω) Per(Ω) = λ 4(Ω) Per(Ω) = λ 5(Ω) Per(Ω) = λ 6(Ω) Per(Ω) = λ 7(Ω) Per(Ω) = λ 8(Ω) Per(Ω) = λ 9(Ω) Per(Ω) = λ 10(Ω) Per(Ω) = Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 5 / 15

16 Examples Main questions in general: Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 6 / 15

17 Examples Main questions in general: Existence of an optimal set? Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 6 / 15

18 Examples Main questions in general: Existence of an optimal set? (among open sets?) Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 6 / 15

19 Examples Main questions in general: Existence of an optimal set? (among open sets?) Regularity of optimal shapes? Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 6 / 15

20 Previous results (non exhaustive!) G = 0: [Gonzalez-Massari-Tamanini 83], Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 7 / 15

21 Previous results (non exhaustive!) G = 0: [Gonzalez-Massari-Tamanini 83], Regularity of quasi-minimizers [Tamanini 88]: Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 7 / 15

22 Previous results (non exhaustive!) G = 0: [Gonzalez-Massari-Tamanini 83], Regularity of quasi-minimizers [Tamanini 88]: Ω is a quasi-minimizer of the perimeter if there exist C R, α (d 1, d] and r 0 > 0 such that for every ball B r with r r 0 P(Ω ) P(Ω) + Cr α, Ω such that Ω Ω B r D, Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 7 / 15

23 Previous results (non exhaustive!) G = 0: [Gonzalez-Massari-Tamanini 83], Regularity of quasi-minimizers [Tamanini 88]: Ω is a quasi-minimizer of the perimeter if there exist C R, α (d 1, d] and r 0 > 0 such that for every ball B r with r r 0 Regularity result: P(Ω ) P(Ω) + Cr α, Ω such that Ω Ω B r D, Ω D is C 1,(α d+1)/2 up to a singular set of dimension less than d 8 (or empty) Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 7 / 15

24 Previous results (non exhaustive!) G = 0: [Gonzalez-Massari-Tamanini 83], Regularity of quasi-minimizers [Tamanini 88]: Ω is a quasi-minimizer of the perimeter if there exist C R, α (d 1, d] and r 0 > 0 such that for every ball B r with r r 0 Regularity result: P(Ω ) P(Ω) + Cr α, Ω such that Ω Ω B r D, Ω D is C 1,(α d+1)/2 up to a singular set of dimension less than d 8 (or empty) G = E f 2007]. with f L and non-negative, or G = λ 1, D bounded: [Landais Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 7 / 15

25 Previous results (non exhaustive!) G = 0: [Gonzalez-Massari-Tamanini 83], Regularity of quasi-minimizers [Tamanini 88]: Ω is a quasi-minimizer of the perimeter if there exist C R, α (d 1, d] and r 0 > 0 such that for every ball B r with r r 0 Regularity result: P(Ω ) P(Ω) + Cr α, Ω such that Ω Ω B r D, Ω D is C 1,(α d+1)/2 up to a singular set of dimension less than d 8 (or empty) G = E f with f L and non-negative, or G = λ 1, D bounded: [Landais 2007]. Partial results if f L p with p (d, ) or if f L with no sign assumption [Landais 2008]. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 7 / 15

26 Previous results (non exhaustive!) G = 0: [Gonzalez-Massari-Tamanini 83], Regularity of quasi-minimizers [Tamanini 88]: Ω is a quasi-minimizer of the perimeter if there exist C R, α (d 1, d] and r 0 > 0 such that for every ball B r with r r 0 Regularity result: P(Ω ) P(Ω) + Cr α, Ω such that Ω Ω B r D, Ω D is C 1,(α d+1)/2 up to a singular set of dimension less than d 8 (or empty) G = E f with f L and non-negative, or G = λ 1, D bounded: [Landais 2007]. Partial results if f L p with p (d, ) or if f L with no sign assumption [Landais 2008]. G = λ k but no volume constraint: [De Philippis-Velichkov 2014]. Use of sub/sup-solutions [Bucur]. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 7 / 15

27 Main result Theorem (De Philippis, L, Pierre, Velichkov 2016) Suppose that D R d is a bounded open set of class C 2 or the entire space D = R d, Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 8 / 15

28 Main result Theorem (De Philippis, L, Pierre, Velichkov 2016) Suppose that D R d is a bounded open set of class C 2 or the entire space D = R d, and G = E f with f L p (D), p (d, ] if D is bounded, and p (d, ) if D = R d ; Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 8 / 15

29 Main result Theorem (De Philippis, L, Pierre, Velichkov 2016) Suppose that D R d is a bounded open set of class C 2 or the entire space D = R d, and G = E f with f L p (D), p (d, ] if D is bounded, and p (d, ) if D = R d ; G = F (λ 1,, λ k ), where F : R k R is increasing in each variable and locally Lipschitz continuous. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 8 / 15

30 Main result Theorem (De Philippis, L, Pierre, Velichkov 2016) Suppose that D R d is a bounded open set of class C 2 or the entire space D = R d, and G = E f with f L p (D), p (d, ] if D is bounded, and p (d, ) if D = R d ; G = F (λ 1,, λ k ), where F : R k R is increasing in each variable and locally Lipschitz continuous. Then there exists a solution of the problem min { P(Ω) + G(Ω), Ω open, Ω D, Ω = m }, Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 8 / 15

31 Main result Theorem (De Philippis, L, Pierre, Velichkov 2016) Suppose that D R d is a bounded open set of class C 2 or the entire space D = R d, and G = E f with f L p (D), p (d, ] if D is bounded, and p (d, ) if D = R d ; G = F (λ 1,, λ k ), where F : R k R is increasing in each variable and locally Lipschitz continuous. Then there exists a solution of the problem min { P(Ω) + G(Ω), Ω open, Ω D, Ω = m }, and every solution Ω is bounded and is a quasi-minimizer of the perimeter with exponent d d/p or d respectively. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 8 / 15

32 Main result Theorem (De Philippis, L, Pierre, Velichkov 2016) Suppose that D R d is a bounded open set of class C 2 or the entire space D = R d, and G = E f with f L p (D), p (d, ] if D is bounded, and p (d, ) if D = R d ; G = F (λ 1,, λ k ), where F : R k R is increasing in each variable and locally Lipschitz continuous. Then there exists a solution of the problem min { P(Ω) + G(Ω), Ω open, Ω D, Ω = m }, and every solution Ω is bounded and is a quasi-minimizer of the perimeter with exponent d d/p or d respectively. Consequence: C 1,β -regularity up to a singular set of dimension less than d 8. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 8 / 15

33 Sharpness of the hypotheses Non-existence situations f L (R d ) such that 0 f < 1 and f (x) x 1, D = R d. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 9 / 15

34 Sharpness of the hypotheses Non-existence situations f L (R d ) such that 0 f < 1 and f (x) x 1, D = R d. Then P(Ω) + E f (Ω) > P(Ω) + E 1 (Ω) P(B) + E 1 (B), Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 9 / 15

35 Sharpness of the hypotheses Non-existence situations f L (R d ) such that 0 f < 1 and f (x) x 1, D = R d. Then P(Ω) + E f (Ω) > P(Ω) + E 1 (Ω) P(B) + E 1 (B), while a sequence of balls of volume m that goes to achieves equality in the limit. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 9 / 15

36 Sharpness of the hypotheses Non-existence situations f L (R d ) such that 0 f < 1 and f (x) x 1, D = R d. Then P(Ω) + E f (Ω) > P(Ω) + E 1 (Ω) P(B) + E 1 (B), while a sequence of balls of volume m that goes to achieves equality in the limit. G = λ 1 and { D = (x, y) (0, ) R, y 2 < x } R 2, and m = B(0, 1). x + 1 Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 9 / 15

37 Sharpness of the hypotheses Non-existence situations f L (R d ) such that 0 f < 1 and f (x) x 1, D = R d. Then P(Ω) + E f (Ω) > P(Ω) + E 1 (Ω) P(B) + E 1 (B), while a sequence of balls of volume m that goes to achieves equality in the limit. G = λ 1 and { D = (x, y) (0, ) R, y 2 < x } R 2, and m = B(0, 1). x + 1 P(Ω) + λ 1 (Ω) > P(B 1 ) + λ 1 (B 1 ), Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 9 / 15

38 Sharpness of the hypotheses Non-existence situations f L (R d ) such that 0 f < 1 and f (x) x 1, D = R d. Then P(Ω) + E f (Ω) > P(Ω) + E 1 (Ω) P(B) + E 1 (B), while a sequence of balls of volume m that goes to achieves equality in the limit. G = λ 1 and { D = (x, y) (0, ) R, y 2 < x } R 2, and m = B(0, 1). x + 1 P(Ω) + λ 1 (Ω) > P(B 1 ) + λ 1 (B 1 ), while equality is achieved for a sequence of sets converging to the ball at infinity. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 9 / 15

39 Elements of proof I Representatives of sets Classes of domains: Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 10 / 15

40 Elements of proof I Representatives of sets Classes of domains: G adapted to open/quasi-open sets Ω w Ω : Ω = {w Ω > 0}, P adapter to measurable sets Ω 1 Ω. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 10 / 15

41 Elements of proof I Representatives of sets Classes of domains: G adapted to open/quasi-open sets Ω w Ω : Ω = {w Ω > 0}, P adapter to measurable sets Ω 1 Ω. Idea: change G to G by replacing the classical space H0 1 (Ω) by H 0 1 (Ω) = { u H 1 (R d ) : u = 0 a.e. on R d \ Ω }. Then one has G G for open sets (strict in general). Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 10 / 15

42 Elements of proof I Representatives of sets Classes of domains: G adapted to open/quasi-open sets Ω w Ω : Ω = {w Ω > 0}, P adapter to measurable sets Ω 1 Ω. Idea: change G to G by replacing the classical space H0 1 (Ω) by H 0 1 (Ω) = { u H 1 (R d ) : u = 0 a.e. on R d \ Ω }. Then one has G G for open sets (strict in general). Regularity of Ω is meant for Ω :={x R d, r > 0, 0 < Ω B r (x) < B r }. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 10 / 15

43 Elements of proof II Existence Solutions of { min P(Ω) + G(Ω), } Ω measurable D Ω = m Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 11 / 15

44 Elements of proof II Existence Solutions of { min P(Ω) + G(Ω), } Ω measurable D Ω = m If D bounded, classical method with sets of finite perimeter. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 11 / 15

45 Elements of proof II Existence Solutions of { min P(Ω) + G(Ω), } Ω measurable D Ω = m If D bounded, classical method with sets of finite perimeter. If D unbounded, concentration compactness (Lions, Bucur) Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 11 / 15

46 Elements of proof II Existence Solutions of { min P(Ω) + G(Ω), } Ω measurable D Ω = m If D bounded, classical method with sets of finite perimeter. If D unbounded, concentration compactness (Lions, Bucur) Compactness (good situation) Compactness at infinity (only if G = E f : hypotheses on f ) Vanishing (easy to exclude) Dichotomy: difficult to exclude. We use boundedness of solutions, which relies on weak regularity theory. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 11 / 15

47 Elements of proof III Penalization of the volume constraint If Ω is solution of { min P(Ω) + G(Ω), } Ω D, Ω = m. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 12 / 15

48 Elements of proof III Penalization of the volume constraint If Ω is solution of { min P(Ω) + G(Ω), } Ω D, Ω = m. then Ω is also solution of (a localized version of) { min P(Ω) + G(Ω)+µ Ω Ω }, Ω D. for some µ R +. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 12 / 15

49 Elements of proof III Penalization of the volume constraint If Ω is solution of { min P(Ω) + G(Ω), } Ω D, Ω = m. then Ω is also solution of (a localized version of) { min P(Ω) + G(Ω)+µ Ω Ω }, Ω D. for some µ R +. General approach based on first order shape derivatives and lipschitz estimates: for Φ W 1, (R d, R d ), G(Φ(Ω)) G(Ω) Cd, Ω,G Φ Id W 1,. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 12 / 15

50 Elements of proof IV Regularity I: Supersolutions for P + µ Assume (thanks to monotonicity of G) Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 13 / 15

51 Elements of proof IV Regularity I: Supersolutions for P + µ Assume (thanks to monotonicity of G) P(Ω ) + µ Ω P(Ω) + µ Ω, for every Ω with Ω Ω. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 13 / 15

52 Elements of proof IV Regularity I: Supersolutions for P + µ Assume (thanks to monotonicity of G) P(Ω ) + µ Ω P(Ω) + µ Ω, for every Ω with Ω Ω. Then H0 1(Ω ) = H 0 1(Ω ), w Ω is locally Lipschitz continuous on R d. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 13 / 15

53 Elements of proof IV Regularity I: Supersolutions for P + µ Assume (thanks to monotonicity of G) P(Ω ) + µ Ω P(Ω) + µ Ω, for every Ω with Ω Ω. Then H0 1(Ω ) = H 0 1(Ω ), w Ω is locally Lipschitz continuous on R d. Based on Exterior estimates: x Ω, B r (x) Ω c B r c < 1, Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 13 / 15

54 Elements of proof IV Regularity I: Supersolutions for P + µ Assume (thanks to monotonicity of G) P(Ω ) + µ Ω P(Ω) + µ Ω, for every Ω with Ω Ω. Then H0 1(Ω ) = H 0 1(Ω ), w Ω is locally Lipschitz continuous on R d. Based on Exterior estimates: x Ω, B r (x) Ω c B r c < 1, Mean curvature bounds in the viscosity sense: [ ] H Ω µ regularity for elliptic PDE. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 13 / 15

55 Elements of proof V Regularity II: Subsolutions Assume P(Ω ) + G(Ω ) + µ Ω P(Ω) + G(Ω) + µ Ω, Ω Ω, and w Ω is lipschitz continuous on R d. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 14 / 15

56 Elements of proof V Regularity II: Subsolutions Assume P(Ω ) + G(Ω ) + µ Ω P(Ω) + G(Ω) + µ Ω, Ω Ω, and w Ω is lipschitz continuous on R d. Then Ω is a quasi-minimizer for interior perturbations. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 14 / 15

57 Elements of proof V Regularity II: Subsolutions Assume P(Ω ) + G(Ω ) + µ Ω P(Ω) + G(Ω) + µ Ω, Ω Ω, and w Ω is lipschitz continuous on R d. Then Ω is a quasi-minimizer for interior perturbations. Based on control variations of G by variations of Ẽ1. Lipschitz continuity of w Ω. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 14 / 15

58 Last remarks Further regularity: classical if G = E f. More involved if G = λ k : see [Bogosel,Oudet 2014] and [Bogosel 2016]. Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 15 / 15

59 Last remarks Further regularity: classical if G = E f. More involved if G = λ k : see [Bogosel,Oudet 2014] and [Bogosel 2016]. Similar problem [van den Berg]: min { λ k (Ω), Ω R d, Ω = m, P(Ω) = p }. Existence if instead of = [Bucur, Mazzoleni 2015]. Regularity? Jimmy Lamboley (Paris-Dauphine) 22/11/16, CIRM 15 / 15