an approximation scheme for an eikonal equation with
play

An approximation scheme for an Eikonal Equation with discontinuous - PowerPoint PPT Presentation

An approximation scheme for an Eikonal Equation with discontinuous coefficient Adriano Festa 1 Maurizio Falcone 2 1 EEE Department, IC London. 2 Dipartimento di Matematica, SAPIENZA - Universita di Roma. HYP 2012 Padova, 25th June - 29th June,


  1. An approximation scheme for an Eikonal Equation with discontinuous coefficient Adriano Festa 1 Maurizio Falcone 2 1 EEE Department, IC London. 2 Dipartimento di Matematica, SAPIENZA - Universita’ di Roma. HYP 2012 Padova, 25th June - 29th June, 2012 Festa-Falcone Discontinuous Eikonal Equation

  2. Outline of the talk Degenerate eikonal equation with discontinuous 1 coefficient Theoretical background 2 A semiLagrangian scheme with good properties 3 Tests and Applications 4 Concluding Remarks 5 Festa-Falcone Discontinuous Eikonal Equation

  3. Degenerate Eikonal Equation with discontinuous coefficient Ω ⊂ R n open bounded domain with regular boundary N  � b ij ( x ) U x i U x j = [ f ( x )] 2  x ∈ Ω  (EK) i , j = 1   U ( x ) = G ( x ) x ∈ ∂ Ω G : ∂ Ω → [ 0 , + ∞ [ is continuous b is symmetric, positive semidefinite kj ) with σ ( · ) : Ω → R NM is L-Lipschitz ( b i , j ) = ( σ ik ) · ( σ t continuous but possibly degenerate f : R N → [ ρ, + ∞ [ , ρ > 0 is Borel measureable but possibly discontinuous Festa-Falcone Discontinuous Eikonal Equation

  4. Optimal Control InterprGammation Rewriting the differential operator in the following form N M ( p · σ k ( x )) 2 = | p · σ ( x ) | 2 , � � b ij ( x ) p i p j = i , j = 1 k = 1 where ( σ ik ) k := σ k : Ω → R N , k = 1 , ... M . Festa-Falcone Discontinuous Eikonal Equation

  5. Optimal Control InterprGammation Rewriting the differential operator in the following form N M ( p · σ k ( x )) 2 = | p · σ ( x ) | 2 , � � b ij ( x ) p i p j = i , j = 1 k = 1 where ( σ ik ) k := σ k : Ω → R N , k = 1 , ... M . We get the equivalent Bellman equation � � � a k σ k ( x ) max − DU ( x ) · = f ( x ) (BL) | a |≤ 1 k = 1 associated to the symmetric optimal control system M � a k σ k ( y ) , y = ˙ y ( 0 ) = x , k = 1 where a : [ 0 , + ∞ [ → { a ∈ R M : | a | ≤ 1 } and y ( · ) ≡ y x ( · , a ) is a solution. Festa-Falcone Discontinuous Eikonal Equation

  6. Viscosity solution via semicontinuous envelopes f ∗ ( x ) = lim r → 0 + inf { f ( y ) : | y − x | ≤ r } , f ∗ ( x ) = lim r → 0 + sup { f ( y ) : | y − x | ≤ r } Definition (Discontinuous Viscosity Solution [Ishii 1985]) A lower semicontinuous (resp. upper) function U is a viscosity super- solution (resp. sub-) of the equation (EK) if for all φ ∈ C 1 (Ω) , and x 0 ∈ argmin x ∈ Ω ( U − φ ) , (resp. x 0 ∈ argmax x ∈ Ω ( U − φ ) ), we have N b ij ( x 0 ) φ x i ( x 0 ) φ x j ( x 0 ) ≥ [ f ∗ ( x 0 )] 2 , � i , j = 1 N b ij ( x 0 ) φ x i ( x 0 ) φ x j ( x 0 ) ≤ [ f ∗ ( x 0 )] 2 ). � (resp. i , j = 1 A function U is a discontinuous viscosity solution of the equation (EK) if U ∗ is a subsolution and U ∗ is a supersolution . Festa-Falcone Discontinuous Eikonal Equation

  7. Dirichelet Conditions Definition (DC in weaker sense) We say that an upper semicontinuous function U , subsolution of the equation in (EK), satisfies weakly the DC if for all φ ∈ C 1 and ˆ x ∈ ∂ Ω , ˆ x ∈ argmax x ∈ Ω ( U − φ ) such that U (ˆ x ) > G (ˆ x ) , then we have N x )] 2 . � b ij (ˆ x ) φ x i φ x j ≤ [ f ∗ (ˆ i , j = 1 Lower semicontinuous functions that satisfy weakly the DC are defined accordingly. Festa-Falcone Discontinuous Eikonal Equation

  8. Hypotheses on the discontinuity interface Assumptions on discontinuities Γ = { x ∈ R N : f is discontinuous at x } is a disjoint union of finite Lipschitz hypersurfaces ( ∃ η + transversal vector) f is continuous in each component Ω ± � � if x ∈ Γ , f ( x ) ∈ Ω − ∋ y → x f ( y ) , lim Ω + ∋ y → x f ( y ) lim . x ∈ Γ ∩ ∂ Ω we can choose η + , η − both inward for Ω . if ˆ Festa-Falcone Discontinuous Eikonal Equation

  9. Comparison Results [Soravia 2006] Theorem Let Ω be an open domain with Lipschitz boundary. Let U , V : Ω → R be upper and a lower- semicontinuous function, a subsolution and a supersolution of (EK) with weak Dirichlet conditions. Suppose that V is nontangentially continuous on ∂ Ω \ Γ in the inward direction η Ω and on Γ ∩ Ω in the direction of η + . Then U ≤ V in Ω . Corollary Let U : Ω → R be a continuous, bounded viscosity solution of the problem (EK) . Then U is unique. Festa-Falcone Discontinuous Eikonal Equation

  10. Example I � � 2 = [ f ( x , y )] 2 x 2 ( u x ( x , y )) 2 + � ] − 1 , 1 [ × ] − 1 , 1 [ u y ( x , y ) u ( ± 1 , y ) = u ( x , ± 1 ) = 0 x , y ∈ [ − 1 , 1 ] where f ( x , y ) = 2, for x > 0, and f ( x , y ) = 1 for x ≤ 0. � x 2 � x � � 0 0 b i , j = , σ ( x ) = , 0 1 0 1 therefore the Bellman’s equation in this case is � − Du ( x , y ) · a 1 ( x , 0 ) T − Du ( x , y ) · a 2 ( 0 , 1 ) T � max = f ( x , y ) . | a |≤ 1 Festa-Falcone Discontinuous Eikonal Equation

  11. Example II  2 ( 1 − | y | ) x > 0 , | y | > 1 + ln x  − 2 ln ( x ) x > 0 , | y | ≤ 1 + ln x u ( x , y ) = (1) u ( − x , y )  x ≤ 0 . 2 is a viscosity solution of the problem. Festa-Falcone Discontinuous Eikonal Equation

  12. A semiLagrangian Approx.: time discretization (Kruzkov’s transform). V ( x ) = 1 − e − U ( x ) 1 | DV ( x ) · σ ( x ) | = f ( x )( 1 − V ( x )) (1 / f as velocity) 2 | DV ( x ) · σ ( x ) | = 1 − V ( x ) f ( x ) (Bellman-type equation) 3 k a k σ k ( x ) �� � sup · DV ( x ) = 1 − V ( x ) f ( x ) a ∈ B ( 0 , 1 ) (discretize as diretional derivative) 4  k a k σ k ( x ) � � �� � 1 h V h ( x ) = inf V h x − h + x ∈ Ω  1 + h f ( x ) 1 + h a ∈ B ( 0 , 1 ) V h ( x ) = 1 − e − G ( x ) x ∈ ∂ Ω  (SDE) Festa-Falcone Discontinuous Eikonal Equation

  13. The set A(x,h) � a k σ k ( x ) � � Figure: The set A ( x , h ) := x − h ; x ∈ B ( 0 , 1 ) in dimension f ( x ) 2. In grey A ( x , h ) := Ω ∩ A ( x , h ) Festa-Falcone Discontinuous Eikonal Equation

  14. A semiLagrangian Approx.: space discretization Let us assume that Ω = Π n i = 1 ( a i , b i ) , Ω ∆ x := Z n ∆ x ∩ Ω and that the grind size ∆ x > 0. We look for a solution of  k a k σ k ( x α ) � 1 h a ∈ B ( 0 , 1 ) I [ W ]( x α − h x α ∈ Ω ∆ x W ( x α ) = min ) +  1 + h f ( x α ) 1 + h W ( x α ) = 1 − e − φ ( x α ) x α ∈ ∂ Ω ∆ x  (SL) where I [ W ]( x ) is a linear interpolation , in the space W ∆ x := � W : Ω → R | W ∈ C (Ω) and DW ( x ) = c α for any x ∈ ( x α , x α + 1 ) } . Festa-Falcone Discontinuous Eikonal Equation

  15. Properties of the scheme (There exists a fixed point ) Let A ( x α , h ) ∈ Ω , for every x α ∈ Ω ∆ x , for any a ∈ B ( 0 , 1 ) , so there exists a unique solution W of (SL) in W ∆ x (Consistency) Developing with che usual Taylor expansion, like in the DF case, we find consistency ( Monotonicity ) The following estimate holds true: � n � 1 || W n − W || ∞ ≤ || W 0 − W || ∞ . 1 + h Festa-Falcone Discontinuous Eikonal Equation

  16. Error estimates Theorem Under the introduced Hypotheses, we have that √ h + C ′ ∆ x || V ( x ) − W ( x ) || L 1 (Ω) ≤ C for all h > 0 h for some constant C , C ′ > 0 independent from h. Moreover, if v ( x ) ∈ C (Ω) we have √ h + 1 + h ( C ′ ∆ x ) || V ( x ) − W ( x ) || L ∞ (Ω) ≤ C for all h > 0 h for some constant C , C ′ > 0 independent from h. Festa-Falcone Discontinuous Eikonal Equation

  17. Sketch of the proof I We start introducing, for a ǫ > 0, the set Λ 2 ǫ := { x ∈ Ω | B ( x , 2 ǫ ) ∩ Λ � = ∅} . We observe that � || V ( x ) − W ( x ) || L 1 (Ω) ≤ | V ( x ) − W ( x ) | dx Ω \ Λ 2 ǫ � � + | V ( x ) − W ( x ) | dx ≤ | V ( x ) − W ( x ) | dx + m (Λ 2 ǫ ) Λ 2 ǫ Ω \ Λ 2 ǫ from the fact that | V ( x ) − W ( x ) | ≤ 1 for all x ∈ Ω . If x ∈ ∂ Ω the assumption is trivially verified because of Dirichlet boundary conditions. Festa-Falcone Discontinuous Eikonal Equation

  18. Sketch of the proof II Let ˆ x ∈ Ω \ Λ 2 ǫ . Consider for ǫ > 0 the auxiliary function ψ ( x , y ) := V ( x ) − W ( y ) − | x − y | 2 x | 2 − | x − ˆ 2 ǫ 2 It is not hard to check that the boundness of v , v ∗ and the upper semicontinuity of ψ , implies the existence of some ( x , y ) in Ω ± (depending on ǫ ) such that for all x , y ∈ Ω ± . ψ ( x , y ) ≥ ψ ( x , y ) √ After some standard calculations, choosing ǫ = h and the boundness of f and σ , we obtain √ V ( x ) − W ( y ) ≤ C h For C suitable positive constants. Festa-Falcone Discontinuous Eikonal Equation

  19. Test 1 Let Ω := ( − 1 , 1 ) × ( 0 , 2 ) and f : Ω → R be defined by  1 x 1 < 0 ,  f ( x 1 , x 2 ) := 3 / 4 x 1 = 0 1 / 2 x 1 > 0  It is not difficult to see that f satisfies our Hypotheses. We can verify that the function 1  2 x 2 , x 1 ≥ 0 , √   2 x 1 + 1 3 − 1 − 2 x 2 , 3 x 2 ≤ x 1 ≤ 0 , u ( x 1 , x 2 ) := √ x 1 < − 1  x 2 , 3 x 2 . √  is a viscosity solution of | Du | = f ( x ) in the sense of our definition. Festa-Falcone Discontinuous Eikonal Equation

  20. Test 1 - Results || · || ∞ || · || 1 ∆ x = h Ord ( L ∞ ) Ord ( L 1 ) 0.2 3.812 e-1 1.821 e-1 0.1 1.734e-1 1.1364 8.112e-2 1.1666 0.05 8.039e-2 1.1095 3.261e-2 1.3148 0.025 4.359e-2 0.8830 1.616e-2 1.0178 0.0125 2.255e-2 0.9509 7.985e-3 1.0271 Festa-Falcone Discontinuous Eikonal Equation

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend