Lipschitz Metrics for Non-smooth evolutions

Like dokumenter
Graphs similar to strongly regular graphs

Existence of resistance forms in some (non self-similar) fractal spaces

UNIVERSITETET I OSLO

Physical origin of the Gouy phase shift by Simin Feng, Herbert G. Winful Opt. Lett. 26, (2001)

Gradient. Masahiro Yamamoto. last update on February 29, 2012 (1) (2) (3) (4) (5)

Numerical Simulation of Shock Waves and Nonlinear PDE

Splitting the differential Riccati equation

Trust region methods: global/local convergence, approximate January methods 24, / 15

Exam in Quantum Mechanics (phys201), 2010, Allowed: Calculator, standard formula book and up to 5 pages of own handwritten notes.

SVM and Complementary Slackness

Solutions #12 ( M. y 3 + cos(x) ) dx + ( sin(y) + z 2) dy + xdz = 3π 4. The surface M is parametrized by σ : [0, 1] [0, 2π] R 3 with.

FYSMEK1110 Eksamensverksted 23. Mai :15-18:00 Oppgave 1 (maks. 45 minutt)

Universitetet i Bergen Det matematisk-naturvitenskapelige fakultet Eksamen i emnet Mat131 - Differensiallikningar I Onsdag 25. mai 2016, kl.

Stationary Phase Monte Carlo Methods

Qi-Wu-Zhang model. 2D Chern insulator. León Martin. 19. November 2015

Generalization of age-structured models in theory and practice

Slope-Intercept Formula

On time splitting for NLS in the semiclassical regime

Mathematics 114Q Integration Practice Problems SOLUTIONS. = 1 8 (x2 +5x) 8 + C. [u = x 2 +5x] = 1 11 (3 x)11 + C. [u =3 x] = 2 (7x + 9)3/2

TFY4170 Fysikk 2 Justin Wells

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Speed Racer Theme. Theme Music: Cartoon: Charles Schultz / Jef Mallett Peanuts / Frazz. September 9, 2011 Physics 131 Prof. E. F.

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Oppgave 1. ( xφ) φ x t, hvis t er substituerbar for x i φ.

Neural Network. Sensors Sorter

Verifiable Secret-Sharing Schemes

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Dynamic Programming Longest Common Subsequence. Class 27

Ma Flerdimensjonal Analyse Øving 1

HØGSKOLEN I NARVIK - SIVILINGENIØRUTDANNINGEN

Unit Relational Algebra 1 1. Relational Algebra 1. Unit 3.3

Eksamen i FY3452 GRAVITASJON OG KOSMOLOGI Lørdag 19. mai :00 13:00

Second Order ODE's (2P) Young Won Lim 7/1/14

UNIVERSITETET I OSLO

Trigonometric Substitution

On the Existence of Strong Solutions to a Fluid Structure Interaction Problem with Navier Boundary Conditions

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Berry Phases of Boundary Gravitons

melting ECMI Modelling week 2008 Modelling and simulation of ice/snow melting Sabrina Wandl - University of Linz Tuomo Mäki-Marttunen - Tampere UT

Oppgave 1. Norges teknisk-naturvitenskapelige universitet NTNU Institutt for fysikk EKSAMEN I: MNFFY 245 INNFØRING I KVANTEMEKANIKK

FYS2140 Kvantefysikk. Løsningsforslag for Oblig 7

On the uniqueness of vanishing viscosity solutions for Riemann problems for polymer flooding

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Estimating Peer Similarity using. Yuval Shavitt, Ela Weinsberg, Udi Weinsberg Tel-Aviv University

Ringvorlesung Biophysik 2016

Eksamen i TMA4190 Mangfoldigheter Onsdag 4 juni, Tid :

Moving Objects. We need to move our objects in 3D space.

UNIVERSITETET I OSLO

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

ECON3120/4120 Mathematics 2, spring 2004 Problem solutions for the seminar on 5 May Old exam problems

GEF2200 Atmosfærefysikk 2017

Level Set methods. Sandra Allaart-Bruin. Level Set methods p.1/24

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO

H 1 -perturbations of Smooth Solutions for a Weakly Dissipative Hyperelastic-rod Wave Equation

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Geodesic flow on the diffeomorphism group of the circle

Satellite Stereo Imagery. Synthetic Aperture Radar. Johnson et al., Geosphere (2014)

EKSAMENSOPPGAVE I FAG TKP 4105

Solution for INF3480 exam spring 2012

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

EKSAMEN I TFY4250 ATOM- OG MOLEKYLFYSIKK og FY2045 Kvantefysikk Tirsdag 13. desember 2005 kl

NO X -chemistry modeling for coal/biomass CFD

UNIVERSITETET I OSLO

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Energy Dissipation in Hybrid Stars. Sophia Han. Washington University

Eksamen i FY3466 KVANTEFELTTEORI II Tirsdag 20. mai :00 13:00

On multigrid methods for the CahnHilliard equation with obstacle potential

Ma Flerdimensjonal Analyse Øving 11

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

5 E Lesson: Solving Monohybrid Punnett Squares with Coding

Kneser hypergraphs. May 21th, CERMICS, Optimisation et Systèmes

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

NORGES TEKNISK-NATURVITENSKAPELIGE UNIVERSITET Side 1 av 5 INSTITUTT FOR ENERGI- OG PROSESSTEKNIKK

On Capacity Planning for Minimum Vulnerability

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO Det matematisk naturvitenskapelige fakultet

Solvability and Regularity for an Elliptic System Prescribing the Curl, Divergence, and Partial Trace of a Vector Field on Sobolev-Class Domains

Namma Kalvi Mathematics (Sample Question Papers only for Practice)

1 Aksiomatisk definisjon av vanlige tallsystemer

PSi Apollo. Technical Presentation

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

EKSAMENSOPPGAVE I SØK3004 VIDEREGÅENDE MATEMATISK ANALYSE ADVANCED MATHEMATICS

Call function of two parameters

Flows and Critical Points

UNIVERSITETET I OSLO

Løsning til deleksamen 2 i SEKY3322 Kybernetikk 3

(ii) x = 0or 4. (ix) x = or 3. (xii) x = or. (vi) x = 4. (xiv) x = Exercise (i) ( x 3) 8. (iii) ( 3q. (iii) 3( x + 2) 16.

Lecture Notes Mek 4320 Hydrodynamic Wave theory

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

Ma Flerdimensjonal Analyse Øving 6

Eksamen i FY3452 GRAVITASJON OG KOSMOLOGI Lørdag 11. august :00 13:00

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

Level-Rebuilt B-Trees

Ole Isak Eira Masters student Arctic agriculture and environmental management. University of Tromsø Sami University College

INF5820 Natural Language Processing - NLP. H2009 Jan Tore Lønning

Motzkin monoids. Micky East. York Semigroup University of York, 5 Aug, 2016

Transkript:

Lipschitz Metrics for Non-smooth evolutions Alberto Bressan Department of Mathematics, Penn State University bressan@math.psu.edu Alberto Bressan (Penn State) Nonsmooth evolutions 1 / 36

Well-posedness for evolution equations d u(t) = A u(t) dt d dt u(t) v(t) κ u(t) v(t) (P1) = u(t) v(t) e κt u(0) v(0) (P) How can we prove well-posedness of the Cauchy problem if (P1) fails? Construct some other weighted distance d (, ) for which (P1) or (P) hold Alberto Bressan (Penn State) Nonsmooth evolutions / 36

Toy model 1: A discontinuous O.D.E. ẋ = f () = { 1 if < 0, if 0, 1 (t) (t) 1 (0) (0) The flow is an isometry w.r.t. the weighted distance d (, y) = y if, y 0 y if, y 0 + y if < 0 < y (Similar to hyperbolic systems of conservation laws) Alberto Bressan (Penn State) Nonsmooth evolutions 3 / 36

(t) 1 (t) (t) 1 (t) ε ε ε t ε t Alberto Bressan (Penn State) Nonsmooth evolutions 4 / 36

Toy model ẋ = (1) Definition: a solution is admissible iff it is strictly increasing For every initial data (0) = 0, the Cauchy problem has a unique admissible solution, depending continuously on the initial data The estimates (P1)-(P) both fail, for the usual distance d(, y) = y The flow is an isometry w.r.t. the weighted distance d (, y) = y 1 ds s (Similar to Hunter-Saton, Camassa-Holm, variational wave equation...) Alberto Bressan (Penn State) Nonsmooth evolutions 5 / 36

An error estimate Theorem. Let S : X [0, [ X be a Lipschitz semigroup on a metric space (X ; d) satisfying d(s tu, S sv) L d(u, v) + L t s u, v X, t, s [0, T ] Then, for every Lipschitz continuous map w : [0, T ] X one has ( ) ( ) T d w(t + h), S h w(t) d w(t ), S w(0) L lim inf T 0 h 0+ h dt = L T 0 [instantaneous error rate at time t] dt w(t) w(t+h) w(t) S h w(t) S w(t+h) T t h S w(t) T t S w(0) T w(0) Alberto Bressan (Penn State) Nonsmooth evolutions 6 / 36

Riemann distance Length of a path γ : [0, 1] M on a Riemann manifold M γ(s) γ (s) b = γ (1) γ = 1 g ij (γ(s)) γ i (s) γ j (s) ds 0 ij a = γ (0) g = g ij = Riemann tensor d (a, b) = { } inf γ ; γ(0) = a, γ(1) = b Toy problem 1: g() = { 1 if < 0 1/ if > 0 Toy problem : g() = 1 Alberto Bressan (Penn State) Nonsmooth evolutions 7 / 36

Finsler metrics Weighted length of a path γ : [0, 1] M in an infinite dimensional manifold: γ(s) b = γ (1) γ = 1 0 γ(s) γ(s) ds a = γ (0) γ (s) T u = tangent space at u M is a Banach space with norm u. For each s [0, 1], the vector γ(s) T γ(s) has norm γ(s) γ(s) Finsler distance = length of shortest path { } d (a, b) = inf γ ; γ(0) = a, γ(1) = b Alberto Bressan (Penn State) Nonsmooth evolutions 8 / 36

Hyperbolic Systems of Conservation Laws u t + f (u) = 0 u = (u 1,..., u n ) R n conserved quantities f = (f 1,..., f n ) : R n R n flues u t + A(u)u = 0 A(u) = Df (u) STRICTLY HYPERBOLIC: each matri A(u) has real distinct eigenvalues λ 1 (u) < λ (u) < < λ n (u) a basis of eigenvectors r 1 (u),..., r n (u) Alberto Bressan (Penn State) Nonsmooth evolutions 9 / 36

(P1) d u(t) ũ(t) L 1 κ u(t) ũ(t) L 1 FAILS! dt shocks in u shocks in u ~ t τ+ h τ u ~ u L 1 τ τ+h t (P3) u(t) ũ(t) L 1 C u(t) ũ(t) L 1 HOLDS for solutions with small total variation Alberto Bressan (Penn State) Nonsmooth evolutions 10 / 36

A space of generalized tangent vectors εv εξ α u ε u α + α εξ α u piecewise Lipschitz with jumps at points 1 < < < N First order perturbation: u ε () = u() + εv() [u( α+) u( α )] χ [α, α+εξ α] + [u( α+) u( α )] χ [α+εξα, ξ α] ξ α>0 ξ α<0 Tangent vector: (v ; ξ 1,..., ξ n ) T u = L 1 (R) R N v = vertical displacement, ξ α = shift in the jump at α Alberto Bressan (Penn State) Nonsmooth evolutions 11 / 36

A weigthed metric on the tangent space εv εξ α u ε u α + α εξ α Standard ( L 1 ) norm on tangent vectors Weighted norm: (v ; ξ 1,..., ξ N ) =. v L 1 +. n (v ; ξ 1,..., ξ n) = i=1 N u( α+) u( α ) ξ α. α=1 W u i () v i () d + v i = i-th component of v along a basis of eigenvectors r 1(u),..., r n(u) N Wk u α () u( α+) u( α ) ξ α. α=1 Assume: u has a shock at α in the family k α Alberto Bressan (Penn State) Nonsmooth evolutions 1 / 36

W u i () = weight given to an i-wave in u( ) at W u i () = 1 + κ 1 A i () + κ Q A i () = total strengths of all waves approaching an i-wave at Q = Glimm interaction potential, 1 << κ 1 << κ Main steps in the analysis: derive a linearized evolution equation for a tangent vector (v, ξ) d (assuming enough regularity) prove that (v(t), ξ(t)) 0 dt u(t) conclude that the semigroup generated by the hyperbolic system of conservation laws is contractive w.r.t. the equivalent distance d (u, ũ) = inf{ γ ; γ(0) = u, γ(1) = ũ} Alberto Bressan (Penn State) Nonsmooth evolutions 13 / 36

u(0) ~ w (0) θ u (0) u(0) θ u(t) ~ γ t u(t) θ w (t) u ~ (T) u(t) w (T) θ θ u (T) d Assume dt (v, ξ)(t) u(t) 0 for every solution u( ) and every corresponding tangent vector (v, ξ). Let γ t : θ u θ (t) be a path in L 1, with tangent vectors d dθ uθ (t) = w θ (t) = (v θ (t), ξ θ (t)) T u(t) Then the path length γ t = 1 0 (v θ, ξ θ )(t) u θ (t) dθ does not increase in time Alberto Bressan (Penn State) Nonsmooth evolutions 14 / 36

The Camassa-Holm Equation ( ) u (CH) u t + + P = 0 P =. 1 ( ) e u + u ( ) u t ( ) u 3 + 3 + u P = u P ( ) ( ) u uu + t u3 = u P 3 E. = 1 [u (t, ) + u (t, ) ] d = const. Natural domain: u(t, ) H 1 (R) Alberto Bressan (Penn State) Nonsmooth evolutions 15 / 36

Multi-peakons solutions u(t, ) = n p i (t) e q i (t) i=1 u(0) u(t) q q 1 (t < T) u(t) u( τ) ( τ > T) solutions can lose regularity: u but remain Hölder continuous: u(t, ) H 1 C Alberto Bressan (Penn State) Nonsmooth evolutions 16 / 36

Continuous dependence? ( ) u Camassa-Holm : u t + + P = 0 d u(t) v(t) L u(t) v(t) dt FAILS! u(t) v(t) C u(0) v(0) FAILS! Alberto Bressan (Penn State) Nonsmooth evolutions 17 / 36

Eample: take v(t, ) = u(t ε, ) u(0) u(t) q q 1 (t < T) u(t) u( τ) ( τ > T) E(t) = [ u (t, ) + u (t, ) ] d constant in time (ecept at t = T ) Alberto Bressan (Penn State) Nonsmooth evolutions 18 / 36

As t T, u(t) and u(t ε) become orthogonal in H 1 u(t ε) u(t) H 1 u H 1 u(t ε) u(t) u (t) u (t ε) Alberto Bressan (Penn State) Nonsmooth evolutions 19 / 36

Equivalent form of the Camassa-Holm equation (A.B. - Adrian Constantin, ARMA 007) (148 citations!?) Energy variable: ξ R, constant along characteristics t y(t, ξ) y t = u(t, y) y(0, ξ) = ȳ(ξ) ȳ(ξ) 0 (1 + ū ) d = ξ U = u t θ = arctan u T P q = (1 + u ) y ξ Alberto Bressan (Penn State) Nonsmooth evolutions 0 / 36

Conservative Solutions u t + (u /) + P = 0 System of ODEs with non-local source terms U = P t θ = (U P) cos θ t 1 θ sin q t = (U 1 ) P sin θ q U(0, ξ) = ū ( ȳ(ξ) ) θ(0, ξ) = arctan ū (ȳ(ξ) ) q(0, ξ) = 1 P(t, ξ) = 1 { ep } ξ θ(s) cos q(s) ds ξ [ ] U (ξ ) cos θ(ξ ) + 1 sin θ(ξ ) q(ξ ) dξ. The global conservative solution can be obtained as the unique fied point of a contractive transformation Alberto Bressan (Penn State) Nonsmooth evolutions 1 / 36

Note: The previous construction yields a family of solutions such that ū n ū H 1 0 = u n (t) u(t) L 0 This is not enough to derive error formulas for approimate solutions, or prove uniqueness Alberto Bressan (Penn State) Nonsmooth evolutions / 36

A distance functional defined by an Optimal Transport problem for conservative, spatially periodic solutions to Camassa-Holm equation (A.B. and M. Fonte Methods & Applications of Analysis, 005) X = R R T T = R/ZZ Given u H 1 (R), Graph(u) = define { (, u(), arctan u () ) ; R} X Alberto Bressan (Penn State) Nonsmooth evolutions 3 / 36

µ u = measure supported on Graph(u), having density 1 + u w.r.t. Lebesgue measure A (, u(), arctan u ()) B a b µ u (A) = b a ( 1 + u () ) d µ u (B) = 0 Alberto Bressan (Penn State) Nonsmooth evolutions 4 / 36

A Constrained Optimal Transportation Problem Given u, v H 1, consider the corresponding measures µ u, µ v Let ψ : R R be an absolutely continuous, increasing map. Move the mass µ u to µ v, from to ψ(). Transportation cost: J ψ (u, v) = 1 0 [distance] [transported mass] + 1 0 [ecess mass] v u +d ψ() ψ () + ψ ()d Alberto Bressan (Penn State) Nonsmooth evolutions 5 / 36

v u +d ψ() ψ () + ψ ()d total ecess mass: (1 + u ()) (1 + v (ψ()))ψ () d Distance functional: J(u, v) = inf ψ Jψ (u, v) u, v H 1 Key inequality: d dt J( u(t), v(t) ) κ J ( u(t), v(t) ) = uniqueness κ is uniformly bounded on bounded subsets of H 1. Alberto Bressan (Penn State) Nonsmooth evolutions 6 / 36

Why a transportation distance is better? u(t) v(t) u(0) v(0) Alberto Bressan (Penn State) Nonsmooth evolutions 7 / 36

The Hunter-Saton Equation ( ) u u t + = 1 4 ( ) u d, u(0, ) = ū(). u (t, ) L (square integrable derivative) Possible gradient catastrophe: (u ) t + u(u ) = u Smooth solutions conserve energy: (u ) t + [ u(u ) ] = 0 Alberto Bressan (Penn State) Nonsmooth evolutions 8 / 36

Removing a singularity (u ) t + u(u ) = u, θ. = arctan u d dt θ(t, (t)) = θ t + uθ = u 1 + u = cos θ 1 θ = arctan u u = tan θ u 1 + u = tan (θ/) 1 + tan (θ/) = θ sin = 1 cos θ Alberto Bressan (Penn State) Nonsmooth evolutions 9 / 36

Odd solutions to Hunter-Saton equation Burgers 0 0 0 Hunter Saton 0 0 0 u t + (u /) = 1 0 u d Alberto Bressan (Penn State) Nonsmooth evolutions 30 / 36

Dissipative solution: Conservative solution: (possibly as a singular measure) 0 u d decreasing in time 0 u d constant in time u 0 u 0 u dissipative u conservative 0 Alberto Bressan (Penn State) Nonsmooth evolutions 31 / 36

Relation with the Camassa-Holm Equation ( ) u u t + + P = 0 P = 1 ( ) e u + u ( e ) Leading approimation = + ( ) u u t + = 1 4 ( ) u d + φ φ = 0 = Hunter-Saton equation Alberto Bressan (Penn State) Nonsmooth evolutions 3 / 36

A Distance Functional for the Hunter-Saton Equation u v +d ψ() ψ () + ψ ()d ψ : R R smooth, increasing determines a transport plan of the measure u onto the measure v, on the graphs of the two functions (, u()) (ψ(), v(ψ())) d (u, v) = minimum cost among all transportation plans Alberto Bressan (Penn State) Nonsmooth evolutions 33 / 36

Transport distance between mass distributions immersed in a fluid Assume: velocity of fluid is Lipschitz continuous with Lipschitz constant ω t=t µ µ µ t=0 µ d (µ t, µ t ) e ωt d (µ 0, µ 0 ) Alberto Bressan (Penn State) Nonsmooth evolutions 34 / 36

Back to Hunter-Saton u t + uu = 1 0 u d, (u ) t + [ u(u ) ] = 0 u u(t,) u(t,) 1 The measure u d on the graph of the solution is transported in the -u plane horizontal speed = u vertical speed = 1 0 u d (constant for each particle) Alberto Bressan (Penn State) Nonsmooth evolutions 35 / 36

Estimates for the transport distance d (u(t), v(t)) e ωt d (u(0), v(0)) - valid for solutions of Hunter-Saton - similar ideas work for Camassa-Holm Alberto Bressan (Penn State) Nonsmooth evolutions 36 / 36