Mathematical aspects of evolving interfaces: lectures given at the C.I.M.-C.I.M.E. joint Euro-summer school held in Madeira, Funchal, Portugal, July 3-9, 2000

Lecture Notes in Mathematics Editors: J.--M. Morel, Cachan F. Takens, Groningen B. Teissier, Paris Subseries: Fondazione C.I.M.E., Firenze Adviser: Pietro Zecca

1812

3 Berlin Heidelberg New York Hong Kong London Milan Paris Tokyo

L. Ambrosio M. Mimura

K. Deckelnick V.A. Solonnikov

G. Dziuk H.M. Soner

Mathematical Aspects of Evolving Interfaces Lectures given at the C.I.M.-C.I.M.E. joint Euro-Summer School held in Madeira, Funchal, Portugal, July 3--9, 2000 Editors: P. Colli J.F. Rodrigues

13

Authors Luigi Ambrosio Scuola Normale Superiore Piazza dei Cavalieri 7 56126 Pisa, Italy e-mail: [email protected] Klaus Deckelnick Institute for Analysis and Numerical Analysis Otto-von-Guericke University of Magdeburg Universit¨atsplatz 2 39106 Magdeburg, Germany e-mail:[email protected] Gerhard Dziuk Institute for Applied Mathematics University of Freiburg Hermann-Herderstr. 10 79104 Freiburg, Germany e-mail: [email protected] Masayasu Mimura Department of Mathematical and Life Sciences Institute of Nonlinear Sciences and Applied Mathematics Graduate School of Sciences Hiroshima University Higashi-Hiroshima 739-8526, Japan

Vsevolod A. Solonnikov Saint-Petersburg Department of the Steklov Mathematical Institute Fontanka 27 191011 St. Petersburg, Russia e-mail: [email protected] Halil Mete Soner Department of Mathematics Koc University Rumeli Fener Yolu Sariyer 80910 Istanbul, Turkey e-mail: [email protected]

Editors Pierluigi Colli Department of Mathematics ‘F. Casorati’ University of Pavia Via Ferrata 1 27100 Pavia, Italy e-mail: [email protected] Jos´e-Francisco Rodrigues CMAF, University of Lisboa Av. Prof. Gama Pinto 2 1649-003 Lisboa, Portugal e-mail: [email protected]

e-mail: [email protected]

Cataloging-in-Publication Data applied for Bibliographic information published by Die Deutsche Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at http://dnb.ddb.de Mathematics Subject Classification (2000): 35-02, 35Bxx, 35QXX, 35K57, 53C44, 76D05, 80A22 ISSN 0075-8434 ISBN 3-540-14033-6 Springer-Verlag Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specif ically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microf ilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer-Verlag. Violations are liable for prosecution under the German Copyright Law. Springer-Verlag Berlin Heidelberg New York a member of BertelsmannSpringer Science + Business Media GmbH http://www.springer.de c Springer-Verlag Berlin Heidelberg 2003
Printed in Germany The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specif ic statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting: Camera-ready TEX output by the authors SPIN: 10922378

41/3142/du - 543210 - Printed on acid-free paper

Preface

The Euro-Summer School on Mathematical Aspects of Evolving Interfaces gathered senior experts and young researchers at the University of Madeira, Funchal, Portugal, in the week July 3-9, 2000. This meeting arose as a joint school of CIM (Centro Internacional de Matem´ atica, Portugal) and CIME (Centro Internazionale Matematico Estivo, Italy). The school was intended to present an advanced introduction and state of the art of recent analytic, modeling and numerical techniques to the mathematical representation and description of moving interfaces. Five complementary courses were delivered and this volume collects the notes of the lectures. Interfaces are geometrical objects modeling free or moving boundaries and arise in a wide range of phase change problems in physical and biological sciences, in particular in material technology and in dynamics of patterns. Especially in the end of last century, the rigorous study of evolving interfaces in a number of applied fields becomes increasingly important, so that the possibility of describing their dynamics through suitable mathematical models became one of the most challenging and interdisciplinary problems in applied mathematics. It was recognized that essential problems related to evolving interfaces can be modelled by means of partial differential equations and systems in domains whose boundary depends on time. In many complicated cases these boundaries are themselves unknown, and correspond, e.g., to a particular level set, or to the discontinuity set, of some physical quantity. In particular, free boundary problems are boundary value problems for differential equations set in a domain where part of the boundary is ”free” and further conditions allow to exclude undeterminacy. Although the first modern work in a free boundary problem was written by Lam´e and Clapeyron in 1831, who considered a simple model for the solidification of a liquid sphere, in the last decades of the XXth century this interdisciplinary field developed tremendously with many new computational demands and new problems from industry and applied sciences, as well as with increasing contributions from Mathematics. Indeed, problems of this sort

VI

Preface

are concerned with several phenomena of high applied interest. Examples include Stefan type problems, where, tipically, the free boundary is the moving interface between liquid and solid, as well as, more general models of phase transitions. Another important example arises in filtration through porous media; here free boundaries occur as fronts between saturated and unsaturated regions. Interesting examples also come from reaction-diffusion models, fluid dynamics, contact mechanics, superconductivity and so on. Several of these problems are also of direct industrial interest, and offer an interesting opportunity of collaboration among theoretical analysts, mathematical physicists and applied scientists. The Madeira school reported on mathematical advances in some theoretical, modeling and numerical issues concerned with dynamics of interfaces and free boundaries. Specifically, the five courses dealt with an assessment of recent results on the optimal transportation problem (L. Ambrosio), the numerical approximation of moving fronts evolving by mean curvature (G. Dziuk), the dynamics of patterns and interfaces in some reaction-diffusion systems with chemical-biological applications (M. Mimura), evolutionary free boundary problems of parabolic type or for Navier-Stokes equations (V.A. Solonnikov), and a variational approach to evolution problems for the GinzburgLandau functional (H.M. Soner). We expect that these lecture notes will be useful not only to experienced readers, to find a detailed description of results and a presentation of techniques, but also to the beginners that aim to learn some of the mathematical aspects behind the different fields.

The editors Pierluigi Colli, Pavia Jos´e Francisco Rodrigues, Lisboa

CIME’s activity is supported by: Ministero dell’Istruzione, dell’Universit` a e della Ricerca; Ministero degli Affari Esteri – Direzione Generale per la Promozione e la Cooperazione – Ufficio V; Istituto Nazionale di Alta Matematica “Francesco Severi”; E.U. under the Training and Mobility of Researchers Programme and UNESCO-ROSTE, Venice Office.

Table of Contents

Lecture Notes on Optimal Transport Problems Luigi Ambrosio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Some elementary examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Optimal transport plans: existence and regularity . . . . . . . . . . . . . . . . . 3 The one dimensional case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 The ODE version of the optimal transport problem . . . . . . . . . . . . . . . 5 The PDE version of the optimal transport problem and the p-laplacian approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Existence of optimal transport maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Regularity and uniqueness of the transport density . . . . . . . . . . . . . . . 8 The Bouchitt´e–Buttazzo mass optimization problem . . . . . . . . . . . . . . 9 Appendix: some measure theoretic results . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29 32 38 41 43 50

Numerical Approximation of Mean Curvature Flow of Graphs and Level Sets Klaus Deckelnick, Gerhard Dziuk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Mean Curvature Flow of Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 The Differential Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Some analysis for the problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Discretization in Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fully discrete schemes and stability . . . . . . . . . . . . . . . . . . . . . . . . . Convergence of the semi-discrete scheme . . . . . . . . . . . . . . . . . . . . 2.4 Convergence of the fully discrete semi implicit scheme . . . . . . . . . 3 Mean Curvature Flow of Level Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Viscosity Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Convergence of the numerical scheme for the level set problem . The semi-discrete scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The fully discrete scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53 53 55 55 55 56 58 58 59 61 63 75 75 76 78 78 78

1 3 5 14 16

VIII

Table of Contents

3.4 Numerical Tests for the Level Set Algorithm . . . . . . . . . . . . . . . . . The regular case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mildly singular case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The highly singular case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The dumb-bell problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81 81 82 83 83 86

Reaction-Diffusion Systems Arising in Biological and Chemical Systems: Application of Singular Limit Procedures Masayasu Mimura . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 What is diffusion? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Discrete diffusion models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Continuous models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Paradox of diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Diffusive patterns and waves - bistable RD equations . . . . . . . . . . . . . 3.1 Scalar bistable RD equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Two component systems of bistable RD equations . . . . . . . . . . . . 4 Resource-consumer RD systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Bacterial colony model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Grey-Scott model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Competition-diffusion systems and singular limit procedures . . . . . . . 5.1 Spatial segregating limits of two competing species model . . . . . 5.2 Dynamics of triple junctions arising in three competing species model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Chemotactic patterns - aggregation and colonies . . . . . . . . . . . . . . . . . . 6.1 Aggregation of Dictyostelium discoideum . . . . . . . . . . . . . . . . . . . . 6.2 New pattern arising in a chemotaxis-diffusion-growth system . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

113 115 116 117 119

Lectures on Evolution Free Boundary Problems: Classical Solutions Vsevolod A. Solonnikov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The model problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Transformation of the problem (1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Proof of Theorem 3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Linear and model problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Lagrangean coordinates and local existence theorems . . . . . . . . . . . . . 8 Proof of Theorem 5.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Scheme of the proof of Theorem 5.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

123 124 128 134 138 142 146 154 162 166 173

89 89 89 90 91 94 95 98 101 101 107 109 111

Table of Contents

Variational and Dynamic Problems for the Ginzburg-Landau Functional Halil Mete Soner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Energy Lower Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Degree and Jacobian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Covering argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Main lower bound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Jacobian and the GL Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Jacobian estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Compactness in two dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Gamma Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Functions of BnV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Gamma limit of I
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Compactness in Higher Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Dynamic Problems: Evolution of Vortex Filaments . . . . . . . . . . . . . . . . 6.1 Energy identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Mean curvature flow and the distance function . . . . . . . . . . . . . . . 6.3 Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

IX

177 177 183 183 184 193 201 201 207 213 215 217 222 225 226 227 227 231

List of Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

Lecture Notes on Optimal Transport Problems Luigi Ambrosio Scuola Normale Superiore Piazza dei Cavalieri 7, I-56126, Pisa, Italy [email protected]

Contents 1. 2. 3. 4. 5. 6. 7. 8. 9.

Some elementary examples Optimal transport plans: existence and regularity The one dimensional case The ODE version of the optimal transport problem The PDE version of the optimal transport problem and the p-laplacian approximation Existence of optimal transport maps Regularity and uniqueness of the transport density The Bouchitt´e–Buttazzo mass optimization problem Appendix: some measure theoretic results

Introduction These notes are devoted to the Monge–Kantorovich optimal transport problem. This problem, in the original formulation of Monge, can be stated as follows: given two distributions with equal masses of a given material g0 (x), g1 (x) (corresponding for instance to an embankment and an excavation), find a transport map ψ which carries the first distribution into the second and minimizes the transport cost
|x − ψ(x)|g0 (x) dx. C(ψ) := X

The condition that the first distribution of mass is carried into the second can be written as

g0 (x) dx = g1 (y) dy ∀B ⊂ X Borel (1) ψ −1 (B)




B

This work has been partially supported by the GNAFA project “Problemi di Monge–Kantorovich e strutture deboli”

L. Ambrosio et al.: LNM 1812, P. Colli and J.F. Rodrigues (Eds.), pp. 1–52, 2003. c Springer-Verlag Berlin Heidelberg 2003


2

Luigi Ambrosio

or, by the change of variables formula, as g1 (ψ(x)) |det∇ψ(x)| = g0 (x)

for Ln -a.e. x ∈ B

if ψ is one to one and sufficiently regular. More generally one can replace the functions g0 , g1 by positive measures f0 , f1 with equal mass, so that (1) reads f1 = ψ# f0 , and replace the euclidean distance by a generic cost function c(x, y), studying the problem
min c(x, ψ(x)) df0 (x). (2) ψ# f0 =f1

X

The infimum of the transport problem leads also to a c-dependent distance between measures with equal mass, known as Kantorovich–Wasserstein distance. The optimal transport problem and the Kantorovich–Wasserstein distance have a very broad range of applications: Fluid Mechanics [7, 8]; Partial Differential Equations [32, 29]; Optimization [13, 14] to quote just a few examples. Moreover, the 1-Wasserstein distance (corresponding to c(x, y) = |x − y| in (2)) is related to the so-called flat distance in Geometric Measure Theory, which plays an important role in its development (see [6], [24], [30], [27], [44]). However, rather than showing specific applications (for which we mainly refer to the Evans survey [21] or to the introduction of [7]), the main aim of the notes is to present the different formulations of the optimal transport problem and to compare them, focussing mainly on the linear case c(x, y) = |x−y|. The main sources for the preparation of the notes have been the papers by Bouchitt´e–Buttazzo [13, 14], Caffarelli–Feldman–McCann [15], Evans–Gangbo [22], Gangbo–McCann [26], Sudakov [42] and Evans [21]. The notes are organized as follows. In Section 1 we discuss some basic examples and in Section 2 we discuss Kantorovich’s generalized solutions, i.e. the transport plans, pointing out the connection between them and the transport maps. Section 3 is entirely devoted to the one dimensional case: in this situation the order structure plays an important role and considerably simplifies the theory. Sections 4 and 5 are devoted to the ODE and PDE based formulations of the optimal transport problem (respectively due to Brenier and Evans–Gangbo); we discuss in particular the role of the so-called transport density and the equivalence of its different representations.  1 Namely, we prove that any transport density µ can be represented as 0 πt# (|y − 1 x|γ) dt, where γ is an optimal planning, as 0 |Et | dt, where Et is the “velocity field” in the ODE formulation, or as the solution of the PDE div(∇µ uµ) = f1 − f0 , with no regularity assumption on f1 , f0 . Moreover, in the same generality we prove convergence of the p-laplacian approximation. In Section 6 we discuss the existence of the optimal transport map, following essentially the original Sudakov approach and filling a gap in his original proof (see also [15, 43]). Section 7 deals with recent results, related to those obtained in [25], on the regularity and the uniqueness of the transport density.

Lecture Notes on Optimal Transport Problems

3

Section 8 is devoted to the connection between the optimal transport problem and the so-called mass optimization problem. Finally, Section 9 contains a self contained list of the measure theoretic results needed in the development of the theory.

Main notation X B(X) Ln Hk Lip(X) Lip1 (X) Σu πt So (X) Sc (X) M(X) M+ (X) M1 (X) |µ| µ+ , µ− f# µ

a compact convex subset of an Euclidean space Rn Borel σ-algebra of X Lebesgue measure in Rn Hausdorff k-dimensional measure in Rn real valued Lipschitz functions defined on X functions in Lip(X) with Lipschitz constant not greater than 1 the set of points where u is not differentiable projections (x, y) → x + t(y − x), t ∈ [0, 1] open segments ] x, y[[ with x, y ∈ X closed segments [ x, y]] with x, y ∈ X, x = y signed Radon measures with finite total variation in X positive and finite Radon measures in X probability measures in X total variation of µ ∈ [M(X)]n positive and negative part of µ ∈ M(X) push forward of µ by f

1 Some elementary examples In this section we discuss some elementary examples that illustrate the kind of phenomena (non existence, non uniqueness) which can occur. The first one shows that optimal transport maps need not exist if the first measure f0 has atoms. Example 1.1 (Non existence). Let f0 = δ0 and f1 = (δ1 + δ−1 )/2. In this case the optimal transport problem has no solution simply because there is no map ψ such that ψ# f0 = f1 . The following two examples deal with the case when the cost function c in X × X is |x − y|, i.e. the euclidean distance between x and y. In this case we will use as a test for optimality the fact that the infimum of the transport problem is always greater than  
u d(f1 − f0 ) : u ∈ Lip1 (X) . (3) sup X

4

Luigi Ambrosio

Indeed,



u d(f1 − f0 ) =

X


u(ψ(x)) − u(x) df0 (x) ≤

X

|ψ(x) − x| df0 (x) X

for any admissible transport ψ. Actually we will prove this lower bound is sharp if f0 has no atom (see (6) and (13)). Our second example shows that in general the solution of the optimal transport problem is not unique. In the one-dimensional case we will obtain (see Theorem 3.1) uniqueness (and existence) in the class of nondecreasing maps. Example 1.2 (Book shifting). Let n ≥ 1 be an integer and f0 = χ[0,n] L1 and f1 = χ[1,n+1] L1 . Then the map ψ(t) = t + 1 is optimal. Indeed, the cost relative to ψ is n and, choosing the 1-Lipschitz function u(t) = t in (3), we obtain that the supremum is at least n, whence the optimality of ψ follows. But since the minimal cost is n, if n > 1 another optimal map ψ is given by  t + n on [0, 1] ψ(t) = t on [1, n]. In the previous example the two transport maps coincide when n = 1; however in this case there is one more (and actually infinitely many) optimal transport map. Example 1.3. Let f0 = χ[0,1] L1 and f1 = χ[1,2] L1 (i.e. n = 1 in the previous example). We have already seen that ψ(t) = t + 1 is optimal. But in this case also the map ψ(t) = 2 − t is optimal as well. In all the previous examples the optimal transport maps ψ satisfy the condition ψ(t) ≥ t. However it is easy to find examples where this does not happen. Example 1.4. Let f0 = χ[−1,1] L1 and f1 = δ−1 + δ1 . In this case the optimal transport map ψ is unique (modulo L1 -negligible sets); it is identically equal to −1 on [−1, 0) and identically equal to 1 on (0, 1]. The verification is left to the reader as an exercise. We conclude this section with some two dimensional examples. Example 1.5. Assume that 2f0 is the sum of the unit Dirac masses at (1, 1) and (0, 0), and that 2f1 is the sum of the unit Dirac masses at (1, 0) and (0, 1). Then the “horizontal” transport and the “vertical” transport are both optimal. Indeed, the cost of these transports is 1 and choosing u(x1 , x2 ) = x1 in (3) we obtain that the infimum of the transport problem is at least 1. Example 1.6. Assume that f1 is the sum of two Dirac masses at A, B ∈ R2 and assume that f0 is supported on the middle axis between them. Then

|x − ψ(x)| df0 (x) = |x − A| df0 (x) X

X

whenever ψ(x) ∈ {A, B}, hence any admissible transport is optimal.

Lecture Notes on Optimal Transport Problems

5

2 Optimal transport plans: existence and regularity In this section we discuss Kantorovich’s approach to the optimal transport problem. His idea has been to look for optimal transport “plans” , i.e. probability measures γ in the product space X × X, rather than optimal transport maps. We will see that this more general viewpoint can be used in several situations to prove that actually optimal transport maps exist (this intermediate passage through a weak formulation of the problems is quite common in PDE and Calculus of Variations). (MK) Let f0 , f1 ∈ M1 (X). We say that a probability measure γ in M1 (X × X) is admissible if its marginals are f0 and f1 , i.e. π0# γ = f0 ,

π1# γ = f1 .

Then, given a Borel cost function c : X × X → [0, ∞], we minimize
I(γ) := c(x, y) dγ(x, y) X×X

among all admissible γ and we denote by Fc (f0 , f1 ) the value of the infimum. We also call an admissible γ a transport plan. Notice also that in Kantorovich’s setting no restriction on f0 or f1 is necessary to produce admissible transport plans: the product measure f0 × f1 is always admissible. In particular the following definition is well posed and produces a family of distances in M1 (X). Definition 2.1 (Kantorovich–Wasserstein distances). Let p ≥ 1 and f0 , f1 ∈ M1 (X). We define the p-Wasserstein distance between f0 and f1 by  Fp (f0 , f1 ) :=

1/p


|x − y|p dγ

min

π0# γ=f0 , π1# γ=f1

.

(4)

X×X

The difference between transport maps and transport plans can be better understood with the following proposition. Proposition 2.1 (Transport plans versus transport maps). Any Borel transport map ψ : X → X induces a transport plan γψ defined by γψ := (Id × ψ)# f0 .

(5)

Conversely, a transport plan γ is induced by a transport map if γ is concentrated on a γ-measurable graph Γ .

6

Luigi Ambrosio

Proof. Let ψ be a transport map. Since π0 ◦(Id×ψ) = Id and π1 ◦(Id×ψ) = ψ, we obtain immediately that π0# γψ = f0 and π1# γψ = ψ# f0 = f1 . Notice also that, by Lusin’s theorem, the graph of ψ is γψ -measurable. Conversely, let Γ ⊂ X × X be a γ-measurable graph on which γ is concentrated and write Γ = {(x, φ(x)) : x ∈ π0 (Γ )} , for some function φ : π0 (Γ ) → X. Let (Kh ) be an increasing sequence of compact subsets of Γ such that γ(Γ \ Kh ) → 0 and notice that  f0 (π0 (Kh )) = γ π0−1 (π0 (Kh )) ≥ γ(Kh ) → 1. Hence, π0 (Γ ) ⊃ ∪h π0 (Kh ) is f0 -measurable and with full measure in X. Moreover, representing γ as γx ⊗ f0 , as in (58) we get
0 = lim γ(X × X \ Kh ) = γx ({y : (x, y) ∈ / ∪h Kh }) df0 (x) h→∞ X
∗ γx (X \ {φ(x)}) df0 (x) ≥ π0 (Γ )

∗

(here denotes the outer integral). Hence γx is the unit Dirac mass at φ(x) for f0 -a.e. x ∈ X. Since
y dγx (y) x → ψ(x) := X

is a Borel map coinciding with φ f0 -a.e., we obtain that γx is the unit Dirac mass at ψ(x) for f0 -a.e. x. For A, B ∈ B(X) we get
γ(A × B) = γx (B) df0 (x) = f0 ({x : (x, ψ(x)) ∈ A × B}) = γψ (A × B) A

and therefore γ = γψ . The existence of optimal transport plans is a straightforward consequence of the w∗ -compactness of probability measures and of the lower semicontinuity of I. Theorem 2.1 (Existence of optimal plans). Assume that c is lower semicontinuous in X × X. Then there exists γ ∈ M1 (X × X) solving (MK). Moreover, if c is continuous and real valued we have
c(x, ψ(x)) df0 (x) (6) min (MK) = inf ψ# f0 =f1

provided f0 has no atom.

X

Lecture Notes on Optimal Transport Problems

7

Proof. Clearly the set of admissible γ’s is closed, bounded and w∗ -compact for the w∗ -convergence of measures (i.e. in the duality with continuous functions in X × X). Hence, it suffices to prove that I(γ) ≤ lim inf I(γh ) h→∞

∗

whenever γh w -converge to γ. This lower semicontinuity property follows by the fact that c can be approximated from below by an increasing sequence of continuous and real valued functions ch (this is a well known fact: see for instance Lemma 1.61 in [4]). The functionals Ih induced by ch converge monotonically to I, whence the lower semicontinuity of I follows. In order to prove the last part we need to show the existence, for any γ ∈ M+ (X × X) with π0# γ = f0 and π1# γ = f1 , of Borel maps ψh : X → X such that ψh# f0 = f1 and δψh (x) ⊗ f0 weakly converge to γ in M(X × X). The approximation Theorem 9.3 provides us, on the other hand, with a Borel map ϕ : X → X such that ϕ# f0 has no atom, is arbitrarily close to f1 and δϕ(x) ⊗ f0 is arbitrarily close (with respect to the weak topology) to γ. We will build ψh by an iterated application of this result. By a standard approximation argument we can assume that L = Lip(c) is finite and that δ|x − y| ≤ c(x, y) ∀x, y ∈ X, for some δ > 0. Possibly replacing c by c/δ, we assume δ = 1. Fix now an integer h, set f00 = f0 and choose ϕ0 : X → X such that ϕ0# f00 has no atom and
F1 (f1 , ϕ0# f00 ) < 2−h , c(ϕ0 (x), x) df00 (x) < Fc (f1 , f00 ) + 2−h . X

Then, we set and

f01

=

ϕ0# f00

and find ϕ1 : X → X such that ϕ1# f01 has no atom

F1 (f1 , ϕ1# f01 ) < 2−1−h ,




c(ϕ1 (x), x) df01 (x) < Fc (f1 , f01 ) + 2−1−h . X

Proceeding inductively and setting f0k = ϕ(k−1)# f0k−1 , we find ϕk such that
c(ϕk (x), x) df0k (x) < Fc (f1 , f0k ) + 2−k−h , F1 (f1 , ϕk# f0k ) < 2−k−h , X

and has no atom. Then, we set φ0 (x) = x and φk = ϕk−1 ◦ · · · ◦ ϕ0 for k ≥ 1, so that f0k = φk# f0 . We claim that (φk ) is a Cauchy sequence in L1 (X, f0 ; X). Indeed, ∞
∞




|φk+1 (x) − φk (x)| df0 (x) = |ϕk (y) − y| df0k (y) ϕk# f0k

k=0

X

k=0

X

≤ 21−h +

∞

k=0

F1 (f1 , f0k ) < ∞.

8

Luigi Ambrosio

Denoting by ψh the limit of φk , passing to the limit as k → ∞ we obtain ψh# f0 = f1 ; moreover, we have
c(φk (x), x) df0 (x)


X

≤

c(ϕ0 (x), x) df0 (x) + L X

k


i=1

≤ Fc (f1 , f0 ) + 2−h + L

k


i=1

|φi (x) − φi−1 (x))| df0 (x)

X

|ϕi (y) − y| df0i (y) ≤ Fc (f1 , f0 ) + 2−h (1 + 2L).

X

Passing to the limit as k → ∞ we obtain
c(ψh (x), x) df0 (x) ≤ Fc (f1 , f0 ) + 2−h (1 + 2L) X

and the proof is achieved. For instance in the case of Example 1.1 (where transport maps do not exist at all) it is easy to check that the unique optimal transport plan is given by 1 1 δ0 × δ−1 + δ0 × δ1 . 2 2 In general, however, uniqueness fails because of the linearity of I and of the convexity of the class of admissible plans γ. In Example 1.2, for instance, any measure t(Id × ψ1 )# f0 + (1 − t)(Id × ψ2 )# f0 is optimal, with t ∈ [0, 1] and ψ1 , ψ2 optimal transport maps. In order to understand the regularity properties of optimal plans γ we introduce, following [26, 36], the concept of cyclical monotonicity. Definition 2.2 (Cyclical monotonicity). Let Γ ⊂ X × X. We say that Γ is c-cyclically monotone if n

i=1

c(xi+1 , yi ) ≥

n

c(xi , yi )

(7)

i=1

whenever n ≥ 2 and (xi , yi ) ∈ Γ for 1 ≤ i ≤ n, with xn+1 = x1 . The cyclical monotonicity property can also be stated in a (apparently) stronger form n n



c(xσ(i) , yi ) ≥ c(xi , yi ) (8) i=1

i=1

Lecture Notes on Optimal Transport Problems

9

for any permutation σ : {1, . . . , n} → {1, . . . , n}. The equivalence can be proved either directly (reducing to the case when σ has no nontrivial invariant set) or verifying, as we will soon do, that any cyclically monotone set is contained in the c-superdifferential of a c-concave function and then checking that the superdifferential fulfils (8). Theorem 2.2 (Regularity of optimal plans). Assume that c is continuous and real valued. Then, for any optimal γ the set spt γ is c-cyclically monotone. Moreover, the union of spt γ as γ range among all optimal plans is c-cyclically monotone. Proof. Assume by contradiction that there exist an integer n ≥ 2 and points (xi , yi ) ∈ spt γ, i = 1, . . . , n, such that f ((xi ), (yi )) :=

n

c(xi+1 , yi ) − c(xi , yi ) < 0,

i=1

with xn+1 = x1 . For 1 ≤ i ≤ n, let Ui , Vi be compact neighbourhoods of xi and yi , respectively, such that γ(Ui × Vi ) > 0 and f ((ui ), (vi )) < 0 whenever ui ∈ Ui and vi ∈ Vi . Set now λ = mini γ(Ui × Vi ) and denote by γi ∈ M1 (Ui × Vi ) the normalized restriction of γ to Ui × Vi . We can find a compact space Y , a probability measure σ in Y , and Borel maps ηi = ui × vi : X → Ui × Vi such that γi = ηi# σ for i = 1, . . . , n (it suffices for instance to define Y as the product of Ui × Vi , so that ηi are the projections on the i-coordinate) and define λ

(ui+1 × vi )# σ − (ui × vi )# σ. n i=1 n

γ  := γ +

Since ληi# σ = λγi ≤ γ we obtain that γ  ∈ M+ (X × X); moreover, it is easy to check that π0# γ  = f0 and π1# γ  = f1 . This leads to a contradiction because


n λ  I(γ ) − I(γ) = c(ui+1 , vi ) − c(ui , vi ) dσ < 0. n Y i=1 In order to show the last part of the statement we notice that the collection of optimal transport plans is w∗ -closed and compact. If (γh )h≥1 is a countable dense set, then

h h 

1 γi spt γi = spt h i=1 i=1 is c-cyclically monotone for any h ≥ 1. Passing to the limit as h → ∞ we obtain that the closure of the union of spt γh is c-cyclically monotone. By the density of (γh ), this closure contains spt γ for any optimal plan γ.

10

Luigi Ambrosio

Next, we relate the c-cyclical monotonicity to suitable concepts (adapted to c) of concavity and superdifferential. Definition 2.3 (c-concavity). We say that a function u : X → R is cconcave if it can be represented as the infimum of a family (ui ) of functions given by ui (x) := c(x, yi ) + ti , for suitable yi ∈ X and ti ∈ R. Remark 2.1. [Linear and quadratic case] In the case when c(x, y) is a symmetric function satisfying the triangle inequality, the notion of c-concavity is equivalent to 1-Lipschitz continuity with respect to the metric dc induced by c. Indeed, given u ∈ Lip1 (X, dc ), the family of functions whose infimum is u is simply {c(x, y) + u(y) : y ∈ X} . In the quadratic case c(x, y) = |x − y|2 /2 a function u is c-concave if and only if u − |x|2 /2 is concave. Indeed, u = inf i c(·, yi ) + ti implies 1 1 u(x) − |x|2 = inf x, −yi  + |yi |2 + ti i 2 2 and therefore the concavity of u − |x|2 /2. Conversely, if v = u − |x|2 /2 is concave, from the well known formula v(x) =

inf

y, p∈∂ + v(y)

v(y) + p, x − y

(here ∂ + v is the superdifferential of v in the sense of convex analysis) we obtain 1 |p − x|2 + c(p, y). u(x) = inf y, −p∈∂ + v(y) 2 Definition 2.4 (c-superdifferential). Let u : X → R be a function. The c-superdifferential ∂c u(x) of u at x ∈ X is defined by ∂c u(x) := {y : u(z) ≤ u(x) + c(z, y) − c(x, y) ∀z ∈ X} .

(9)

The following theorem ([40], [41], [36]) shows that the graphs of superdifferentials of c-concave functions are maximal (with respect to set inclusion) c-cyclically monotone sets. It may considered as the extension of the well known result of Rockafellar to this setting. Theorem 2.3. Any c-cyclically monotone set Γ is contained in the graph of the c-superdifferential of a c-concave function. Conversely, the graph of the c-superdifferential of a c-concave function is c-cyclically monotone.

Lecture Notes on Optimal Transport Problems

11

Proof. This proof is taken from [36]. We fix (x0 , y0 ) ∈ Γ and define u(x) := inf c(x, yn ) − c(xn , yn ) + · · · + c(x1 , y0 ) − c(x0 , y0 )

∀x ∈ X,

where the infimum runs among all collections (xi , yi ) ∈ Γ with 1 ≤ i ≤ n and n ≥ 1. Then u is c-concave by construction and the cyclical monotonicity of Γ gives u(x0 ) = 0 (the minimum is achieved with n = 1 and (x1 , y1 ) = (x0 , y0 )). We will prove the inequality u(x) ≤ u(x ) + c(x, y  ) − c(x , y  )

(10)

for any x ∈ X and (x , y  ) ∈ Γ . In particular (choosing x = x0 ) this implies that u(x ) > −∞ and y  ∈ ∂c u(x ). In order to prove (10), we fix λ > u(x ) and find (xi , yi ) ∈ Γ , 1 ≤ i ≤ n, such that c(x , yn ) − c(xn , yn ) + · · · + c(x1 , y0 ) − c(x0 , y0 ) < λ. Then, setting (xn+1 , yn+1 ) = (x , y  ), we find u(x) ≤ c(x, yn+1 ) − c(xn+1 , yn+1 ) + c(xn+1 , yn ) − c(xn , yn ) + · · · + c(x1 , y0 ) − c(x0 , y0 ) ≤ c(x, y  ) − c(x , y  ) + λ. Since λ is arbitrary, (10) follows. Finally, if v is c-concave, yi ∈ ∂c v(xi ) for 1 ≤ i ≤ n and σ is a permutation we can add the inequalities v(xσ(i) ) − v(xi ) ≤ c(xσ(i) , yi ) − c(xi , yi ) to obtain (8). In the following corollary we assume that the cost function is symmetric, continuous and satisfies the triangle inequality, so that c-concavity reduces to 1-Lipschitz continuity with respect to the distance induced by c. Corollary 2.1 (Linear case). Let γ ∈ M1 (X × X) with π0# γ = f0 and π1# γ = f1 . Then γ is optimal for (MK) if and only if there exists u : X → R such that |u(x) − u(y)| ≤ c(x, y) u(x) − u(y) = c(x, y)

∀(x, y) ∈ X × X, for (x, y) ∈ spt γ.

(11) (12)

In addition, there exists u satisfying (11) and such that (12) holds for any optimal planning γ. We will call any function u with these properties a maximal Kantorovich potential. Proof. (Sufficiency) Let γ  be any admissible transport plan; by applying (11) first and then (12), we get

12

Luigi Ambrosio

I(γ  ) ≥




(u(x) − u(y)) dγ  =




X
X×X = (u(x) − u(y)) dγ = I(γ).


u df0 −

u df1 X

X

(Necessity) Let Γ be the closure of the union of spt γ  , as γ  varies among all optimal plans for (MK). Then, we know that Γ is c-cyclically monotone, hence there exists a c-concave function u such that Γ ⊂ ∂c u. Then, (11) follows by the c-concavity of u, while the inclusion Γ ⊂ ∂c u implies u(y) − u(x) ≤ c(y, y) − c(x, y) = −c(x, y) for any (x, y) ∈ spt γ ⊂ Γ . This, taking into account (11), proves (12). A direct consequence of the proof of sufficiency is the identity 
 min (MK) = max ud(f0 − f1 ) : u ∈ Lip1 (X, dc )

(13)

(where dc is the distance in X induced by c) and the maximum on the right is achieved precisely whenever u satisfies (12). In the following corollary, instead, we consider the case when c(x, y) = |x − y|2 /2. The result below, taken from [26], was proved first by Brenier in [9, 10] under more restrictive assumptions on f0 , f1 (see also [26] for the general case c(x, y) = h(|x − y|)). Before stating the result, we recall that the set Σv of points of nondifferentiability of a real valued concave function v (i.e. the set of points x such that ∂ + v(x) is not a singleton) is countably (CC) regular (see [45] and also [1]). This means that Σv can be covered with a countable family of (CC) hypersurface, i.e. graphs of differences of convex functions of n − 1 variables. This property is stronger than the canonical Hn−1 -rectifiability: it implies that Hn−1 -almost all of Σv can be covered by a sequence of C 2 hypersurfaces. Corollary 2.2 (Quadratic case). Assume that any (CC) hypersurface is f0 -negligible. Then the optimal planning γ is unique and is induced by an optimal transport map ψ. Moreover ψ is the gradient of a convex function. Proof. Let u : X → R be a c-concave function such that the graph Γ of its superdifferential contains the support of any optimal planning γ. As c(x, y) = |x − y|2 /2, an elementary computation shows that (x0 , y0 ) ∈ Γ if and only if −y0 ∈ ∂ + v(x0 ), where v(x) = u(x) − |x|2 /2 is the concave function already considered in Remark 2.1. Then, by the above mentioned results on differentiability of concave functions, the set of points where v is not differentiable is f0 -negligible. Hence, for f0 -a.e. x ∈ X, there is a unique y0 = −∇v(x0 ) ∈ X such that (x0 , y0 ) ∈ Γ . As spt γ ⊂ Γ , by Proposition 2.1 we infer that

Lecture Notes on Optimal Transport Problems

13

γ = (Id × ψ)# f0 for any optimal planning γ, with ψ = −∇v. Example 2.1 (Brenier polar factorization theorem). A remarkable consequence of Corollary 2.2 is the following result, known as polar factorization theorem. A vector field r : X → X such that Ln (r−1 (B)) = 0

whenever

Ln (B) = 0

(14)

can be written as ∇u ◦ η, with u convex and η measure preserving. It suffices to apply Corollary 2.2 to the measures f0 = r# (Ln X) (absolutely continuous, due to (14)) and f1 = Ln X; we have then Ln

X = (∇v)# (r# (Ln

X)) = (∇v ◦ r)# (Ln

X),

for a suitable convex function v, hence η = (∇v) ◦ r is measure preserving. The desired representation follows with u = v ∗ , since ∇u = (∇v)−1 . Let us assume that f0 = Ln X and let f1 ∈ M+ (X) be any other measure such that f1 (X) = Ln (X); in general the problem of mapping f0 into f1 through a Lipschitz map has no solution (it suffices to consider, for instance, the case when X = B 1 and f1 = n1 H1 ∂B1 ). However, another remarkable consequence of Corollary 2.2 is that the problem has solution if we require the transport map to be only a function of bounded variation: indeed, bounded monotone functions (in particular gradients of Lipschitz convex functions) are functions of bounded variation (see for instance Proposition 5.1 of [2]). Moreover, we can give a sharp quantitative estimate of the error made in the approximation by Lipschitz transport maps. Theorem 2.4. There exists a constant C = C(n, X) with the following property: for any µ ∈ M+ (X) with µ(X) = Ln (X) and any M > 0 there exist a Lipschitz function φ : X → X and B ∈ B(X) such that Lip(φ) ≤ M , Ln (X \ B) ≤ C/M and µ = φ# (Ln

B) + µs ,

with µs ∈ M+ (X) and µs (X) ≤ Ln (X \ B). Proof. By Corollary 2.2 we can represent µ = ψ# (Ln X) with ψ : X → X equal to the gradient of a convex function. Let Ω be the interior of X; by applying Proposition 5.1 of [2] (valid, more generally, for monotone operators) we obtain that the total variation |Dψ|(Ω) can be estimated with a suitable constant C depending only on n and X. Therefore, by applying Theorem 5.34 of [4] we can find a Borel set B ⊂ X (it is a suitable sublevel set of the maximal function of |Dψ|) such that Ln (X \ B) ≤ c(n)|Dψ|(Ω)/M and the restriction of ψ to B is a M -Lipschitz function, i.e. with Lipschitz constant

14

Luigi Ambrosio

not greater than M . By Kirszbraun theorem (see for instance [24]) we can extend ψ|B to a M -Lipschitz function φ : X → X. Setting B c = X \ B we have then µ = ψ# (Ln

B) + ψ# (Ln

and, setting µs = ψ# (Ln

B c ) = φ# (Ln

B) + ψ# (Ln

Bc)

B c ), the proof is achieved.

3 The one dimensional case In this section we assume that X = I is a closed interval of the real line; we also assume for simplicity that the transport cost is c(x, y) = |x − y|p for some p ≥ 1. Theorem 3.1 (Existence and uniqueness). Assume that f0 is a diffuse measure, i.e. f0 ({t}) = 0 for any t ∈ I. Then (i) there exists a unique (modulo countable sets) nondecreasing function ψ : spt f0 → X such that ψ# f0 = f1 ; (ii) the function ψ in (i) is an optimal transport, and, if p > 1, it is the unique optimal transport. In the one-dimensional case these results are sharp: we have already seen that transport maps need not exist if f0 has atoms (Example 1.1) and that, without the monotonicity constraint, are not necessarily unique when p = 1. Proof. (i) Let m = min I and define ψ(s) := sup {t ∈ I : f1 ([m, t]) ≤ f0 ([m, s])} .

(15)

It is easy to check that the following properties hold: (a) ψ is non decreasing; (b) ψ(I) ⊃ spt f1 ; (c) if ψ(s) is not an atom of f1 , we have f1 ([m, ψ(s)]) = f0 ([m, s]).

(16)

Let T be the at most countable set made by the atoms of f1 and by the points t ∈ I such that ψ −1 (t) contains more than one point; then ψ −1 is well defined on ψ(I) \ T and ψ −1 ([t, t ]) = [ψ −1 (t), ψ −1 (t )] whenever t, t ∈ ψ(I) \ T with t < t . Then (c) gives f1 ([t, t ]) = f1 ([m, t ]) − f1 ([m, t]) = f0 ([m, ψ −1 (t )]) − f0 ([m, ψ −1 (t)]) = f0 ([ψ −1 (t), ψ −1 (t )]) = f0 (ψ −1 ([t, t ]))

Lecture Notes on Optimal Transport Problems

15

(notice that only here we use the fact that f0 is diffuse). By (b) the closed intervals whose endpoints belong to ψ(I) \ T generate the Borel σ-algebra of spt f1 , and this proves that ψ# f0 = f1 . Let φ be any nondecreasing function such that φ# f0 = f1 and assume, possibly modifying φ on a countable set, that φ is right continuous. Let T := {s ∈ spt f0 : (s, s ) ∩ spt f0 = ∅ for some s > s } and notice that T is at most countable (since we can index with T a family of pairwise disjoint open intervals). We claim that φ ≥ ψ on spt f0 \ T ; indeed, for s ∈ spt f0 \ T and s > s we have the inequalities f1 ([m, φ(s )]) = f0 (φ−1 ([m, φ(s )])) ≥ f0 ([m, s ]) > f0 ([m, s]) and, by the definition of ψ, the inequality follows letting s ↓ s. In particular




φ − ψ df0 = φ df0 − ψ df0 = 1 df1 − 1 df1 = 0, I

I

I

I

I

whence φ = ψ f0 -a.e. in I. It follows that φ(s) = ψ(s) at any continuity point s ∈ spt f0 of φ and ψ. (ii) By a continuity argument it suffices to prove that ψ is the unique solution of the transport problem for any p > 1 (see also [26, 15]). Let γ be an optimal planning and notice that the cyclical monotonicity proved in Theorem 2.2 gives |x − y  |p + |x − y|p ≥ |x − y|p + |x − y  |p whenever (x, y), (x , y  ) ∈ spt γ. If x < x , this condition implies that y ≤ y  (this is a simple analytic calculation that we omit, and here the fact that p > 1 plays a role). This means that the set T := {x ∈ spt f0 : card({y : (x, y) ∈ spt γ}) > 1} is at most countable (since we can index with T a family of pairwise disjoint open intervals) hence f0 -negligible. Therefore, for f0 -a.e. x ∈ I, there exists ˜ a unique y = ψ(x) ∈ I such that (x, y) ∈ spt γ (the existence of at least one y follows by the fact that the projection of spt γ on the first factor is spt f0 ). Notice also that ψ˜ is nondecreasing in its domain. Arguing as in Proposition 2.1, we obtain that ˜ # f0 . γ = (Id × ψ) In particular,   ˜ # f0 = ψ˜# f0 f1 = π1# γ = π1# (Id × ψ) and, since ψ˜ is non decreasing, it follows that ψ˜ = ψ (up to countable sets) on spt f0 .

16

Luigi Ambrosio

4 The ODE version of the optimal transport problem In this and in the next section we rephrase the optimal transport problem in differential terms. In the following we consider a fixed auxiliary open set Ω containing X; we assume that Ω is sufficiently large, namely that the open r-neighbourhood of X is contained in Ω, with r > diam(X) (the necessity of this condition will be discussed later on). The first idea, due to Brenier ([8, 7] and also [11]) is to look for all the paths ft in M+ (X) connecting f0 to f1 . In the simplest case when ft = δx(t) , ˙ it turns out that the velocity field Et = x(t)δ x(t) is related to ft by the equation f˙t + ∇ · Et = 0 in (0, 1) × Ω (17) in the distribution sense. Indeed, given ϕ ∈ Cc∞ (0, 1) and φ ∈ Cc∞ (Ω), it suffices to take ϕ(t)φ(x) as test function in (17) and to use the definitions of ft and Et to obtain
1 ϕ(t)φ(x(t)) ˙ + ϕ(t)∇φ(x(t)), x(t) ˙ dt (f˙t + ∇ · Et )(ϕφ) = −


0 1

=− 0

d [ϕ(t)φ(x(t))] dt = 0. dt

More generally, regardless of any assumption on (ft , Et ) ∈ M+ (X) × [M(Ω)]n , it is easy to check that (17) holds in the distribution sense if and only if f˙t (φ) = ∇φ · Et in (0,1) ∀φ ∈ Cc∞ (Ω) (18) in the distribution sense. We will use both interpretations in the following. One more interpretation of (17) is given in the following proposition. Recall that a map f defined in (0, 1) with values in a metric space (E, d) is said to be absolutely continuous if for any ε > 0 there exists δ > 0 such that



(yi − xi ) < δ =⇒ d (f (yi ), f (xi )) < ε i

i

for any family of pairwise disjoint intervals (xi , yi ) ⊂ (0, 1). Proposition 4.1. If some family (ft ) ⊂ M1 (X) fulfils (17) for suitable mea1 sures Et ∈ [M(Ω)]n satisfying 0 |Et |(Ω) dt < ∞, then f is an absolutely continuous map between (0, 1) and M1 (X), endowed with the 1-Wasserstein distance (4) and F1 (ft+h , ft ) ≤ |Et |(Ω) h→0 |h| lim

for L1 -a.e. t ∈ (0, 1).

(19)

Conversely, if ft is an absolutely continuous map, we can choose Et so that equality holds in (19).

Lecture Notes on Optimal Transport Problems

17

Proof. In (13) we can obviosuly restrict to test functions u such that |u| ≤ r = diam(X) on X; by our assumption on Ω any of these function can be extended to Rn in such a way that the Lipschitz constant is still less than 1 and u ≡ 0 in a neighbourhood of Rn \ Ω. In particular, choosing an optimal u and setting uε = u ∗ ρε , we get

t uε d(ft − fs ) = lim ∇ · Eτ (uε ) dτ F1 (fs , ft ) = lim+ ε→0 ε→0+ s X
t
t Eτ · ∇uε dτ ≤ |Eτ |(Ω) dτ (20) = − lim ε→0+

s

s

whenever 0 ≤ s ≤ t ≤ 1 and this easily leads to (19). In the proof of the converse implication we can assume with no loss of generality (up to a reparameterization by arclength) that f is a Lipschitz map. We can consider M1 (X) as a subset of the dual Y = G∗ , where   G := φ ∈ C 1 (Rn ) ∩ Lip(Rn ) : φ ≡ 0 on Rn \ Ω endowed with the norm φ = Lip(φ). By using convolutions and (13), it is easy to check that F1 (µ, ν) = µ − νY , so that M1 (X) is isometrically embedded in Y . Moreover, using Hahn–Banach theorem it is easy to check that any y ∈ Y is representable as the divergence of a measure E ∈ [M(Ω)]n with y = |E|(Ω) (E is not unique, of course). By a general result proved in [5], valid also for more than one independent variable, any Lipschitz map f from (0, 1) into the dual Y of a separable Banach space is weakly∗ -differentiable for L1 -almost every t, i.e. ∃ w∗ − lim

h→0

ft+h − ft =: f˙(t) h

t and ft −fs = s f˙(τ ) dτ for s, t ∈ (0, 1). In addition, the map is also metrically differentiable for L1 -almost every t, i.e. ∃ lim

h→0

ft+h − ft  =: mf˙(t). |h|

Although f˙ is only a w∗ -limit of the difference quotients, it turns out (see [5]) that the metric derivative mf˙ is L1 -a.e. equal to f˙. Putting together these informations the conclusion follows. According to Brenier, we can formulate the optimal transport problem as follows. (ODE) Let f0 , f1 ∈ M1 (X) be given probability measures. Minimize
1 |Et |(Ω) dt (21) J(E) := 0

18

Luigi Ambrosio

among all Borel maps ft : [0, 1] → M+ (X) and Et : [0, 1] → [M(Ω)]n such that (17) holds.

Example 4.1. In the case considered in Example 1.3 the measures ft = χ[t,t+1] L1 ,

Et = χ[t,t+1] L1

provide an admissible and optimal flow, obviously related to the optimal transport map x → x + 1. But, quite surprisingly, we can also define an optimal flow by  1 1   1−2t χ[2t,1] L if 0 ≤ t < 1/2 ft = δ1 , if t = 1/2   1 1 χ L if 1/2 < t ≤ 1 2t−1 [1,2t] whose “velocity field” is  2(1−x) Et =

1 (1−2t)2 χ[2t,1] L 2(x−1) 1 (2t−1)2 χ[1,2t] L

if 0 ≤ t < 1/2 if 1/2 < t ≤ 1

.

It is easy to check that f˙t = ∇ · Et = 0 and that |Et | are probability measures for any t = 1/2, hence J(E) = 1. The relation of this new flow with the optimal transport map x → 2 − x will be seen in the following (see Remark 4.1(3)). In order to relate solutions of (ODE) to solutions of (MK), we will need an uniqueness theorem, under regularity assumptions in the space variable, for the ODE f˙t + ∇ · (gt ft ) = 0. If ft , gt are smooth (say C 2 ) with respect to both the space and time variables, uniqueness is a consequence of the classical method of characteristics (see for instance §3.2 of [20]), which provides the representation 
 t

ft (xt ) = f0 (x) exp −

cs (xs ) ds , 0

where ct = ∇ · gt and xt solves the ODE x˙ t = gt (xt ),

x0 = x,

t ∈ (0, 1).

See also [33] for more general uniqueness and representation results in a weak setting. Theorem 4.1. Assume that f˙t + ∇ · (gt ft ) = 0 1

in (0, 1) × Rn ,

(22)

where 0 |ft |(Rn ) dt < ∞ and |gt | + Lip(gt ) ≤ C, with C independent of t and f0 = 0. Then ft = 0 for any t ∈ (0, 1).

Lecture Notes on Optimal Transport Problems

19

Proof. Let gtε be obtained from gt by a mollification with respect to the space and time variables and define X ε (s, t, x) as the solution of the ODE x˙ = gsε (x) (with s as independent variable) such that X ε (t, t, x) = x. Define, for ψ ∈ Cc∞ ((0, 1) × Rn ) fixed,
ϕε (t, x) := −

1

ψ (s, X ε (s, t, x)) ds. t

Since X ε (s, t, X ε (t, 0, x)) = X ε (s, 0, x), we have


1

ϕε (t, X ε (t, 0, x)) = −

ψ (s, X ε (s, 0, x)) ds t

and, differentiating both sides, we infer  ε  ∂ϕ ε ε + gt · ∇ϕ (t, X ε (t, 0, x)) = ψ (t, X ε (t, 0, x)) ∂t whence ψ = (∂ϕε /∂t + gtε · ∇ϕε ) in (0, 1) × Rn . Insert now the test function ϕε in (22) and take into account that f0 = 0 to obtain
1

1
∂ϕε ε + gt · ∇ϕ ) dxdt = 0= ft ( ft ψ + ft (gt − gtε ) · ∇ϕε dxdt. ∂t n n R R 0 0 The proof is finished letting ε → 0+ and noticing that |gtε | + |∇ϕε | ≤ C, with C independent of ε. In the following theorem we show that (MK) and (ODE) are basically equivalent. Here the assumption that Ω is large enough plays a role: indeed, if for instance X = [ x0 , x1 ] , f0 = δx0 , and f1 = δx1 , the infimum of (ODE) is easily seen to be less than dist(x0 , ∂Ω) + dist(x1 , ∂Ω). Therefore, in the general case when f0 , f1 are arbitrary measures in X, we require that dist(∂Ω, X) > diam(X). Theorem 4.2 ((MK) versus (ODE)). The problem (ODE) has at least one solution and min (ODE) = min (MK). Moreover, for any optimal planning γ ∈ M1 (X × X) for (MK) the measures ft := πt# γ,

Et := πt# ((y − x)γ)

t ∈ [0, 1],

(23)

with πt (x, t) = x + t(y − x), solve (ODE). Proof. Let (ft , Et ) be defined by (23). For any φ ∈ C ∞ (Rn ) we compute

20

Luigi Ambrosio

d dt


φ dft = X

d dt



φ (x + t(y − x)) dγ X×X

∇φ (x + t(y − x)) · (y − x) dγ

= =

X×X n


i=1

∇i φ dEt,i = −∇ · Et (φ);

X

hence the ODE (17) is satisfied. Then, we simply evaluate the energy J(E) in (21) by





1

|πt# ((y − x)γ)|(X) dt ≤

J(E) = 0




1

1

πt# (|y − x|γ)(X) dt

(24)

0




|x − y| dγdt = I(γ).

= 0

X×X

This shows that inf (ODE) ≤ min (MK). In order to prove the opposite inequality we first use Proposition 4.1 and then we present a different strategy, which provides more geometric informations. By Proposition 4.1 we have
1
1 d F1 (ft , f0 ) dt ≤ |Et |(Ω) dt F1 (f0 , f1 ) ≤ 0 dt 0 for any admissible flow (ft , Et ). This proves that min (MK) ≤ inf (ODE). For given (and admissible) (ft , Et ), assuming that


1

|Et | dt ⊂⊂ Ω,

spt 0

we exhibit an optimal planning γ with I(γ) ≤ J(E). To this aim we fix a cut-off function θ ∈ Cc∞ (Ω) with 0 ≤ θ ≤ 1 and θ ≡ 1 1 on X ∪ spt 0 |Et | dt; then, we define ftε := ft ∗ ρε + εθ

and Etε := Et ∗ ρε ,

where ρ is any convolution kernel with compact support and ρε (x) = ε−n ρ(x/ε). Notice that ftε are strictly positive on spt Et ; moreover (17) still holds, spt Etε ⊂ Ω for ε small enough and L1 -a.e. t, and by (53) we have
|Etε |(y) dy ≤ |Et |(Ω) ∀t ∈ [0, 1]. (25) Rn

We define gtε = Etε /ftε and denote by ψtε (x) the semigroup in [0, 1] associated to the ODE

Lecture Notes on Optimal Transport Problems

ψ˙ tε (x) = gtε (ψtε (x)) ,

ψ0ε (x) = x.

21

(26)

Notice that this flow for ε small enough maps Ω into itself and leaves Rn \ Ω ε fixed. Now, we claim that ftε = ψt# (f0ε Ln ) for any t ∈ [0, 1]. Indeed, since by definition f˙tε + ∇ · (gtε ftε ) = f˙tε + ∇ · Etε = 0, by Theorem 4.1 and the linearity of the equation we need only to check ε that also νtε = ψt# (f0ε Ln ) satisfies the ODE ν˙ tε + ∇ · (g ε νtε ) = 0. This is a straightforward computation based on (26). Using this representation, we can view ψ1ε as approximate solutions of the optimal transport problem and define γ ε := (Id × ψ1ε )# (f0ε Ln ),





i.e.

ϕ(x, y) dγ ε (x) =

ϕ (x, ψ1ε (x)) f0ε (x) dx

Rn

for any bounded Borel function ϕ in Rn × Rn . In order to evaluate I(γ ε ), we notice that


1

|gtε |(ψtε (x)) dt;

Length(ψtε (x)) = 0

hence, by multiplying by

f0ε

and integrating we get



Rn

1

Length(ψtε (x))f0ε (x) dx =

Rn

0







1




= 0

Rn




As I(γ ε ) =

Rn

|gtε (y)|ftε (y) dydt
|Etε |(y) dydt

≤

(27) 1

|Et |(Ω) dt. 0


|ψ1ε (x) − x|f0ε (x) dx ≤

Rn

Length(ψtε (x)) df0 (x),

this proves that I(γ ε ) ≤ J(E). Passing to the limit as ε ↓ 0 and noticing that π0# γ ε = f0ε Ln ,

π1# γ ε = f1ε Ln ,

we obtain, possibly passing to a subsequence, an admissible planning γ ∈ M1 (Rn × Rn ) for f0 and f1 such that I(γ) ≤ J(E). We can turn γ in a planning supported in X × X simply replacing γ by (π × π)# γ, where π : Rn → X is the orthogonal projection. Remark 4.1. (1) Starting from (ft , Et ), the (possibly multivalued) operation leading to γ and then to the new flow f˜t := πt# γ,

˜t := πt# ((y − x)γ) E

22

Luigi Ambrosio

can be understood as a sort of “arclength reparameterization” of (ft , Et ). However, since this operation is local in space (consider for instance the case of two line paths, one parameterized by arclength, one not), in general there is no function ϕ(t) such that ˜t ) = (fϕ(t) , Eϕ(t) ). (f˜t , E (2) The solutions (ft , Et ) in Example 4.1 are built as πt# γ and πt# ((y − x)γ) where γ = (Id × ψ)# f0 and ψ(x) = x + 1, ψ(x) = 2 − x. In particular, in the second example, f1/2 = δ1 because 1 is the midpoint of any transport ray. (3) By making a regularization first in space and then in time of (ft , Et ) one can avoid the use of Theorem 4.1, using only the classical representation of solutions of (22) with characteristics. Remark 4.2. [Optimality conditions] (1) If (ft , Et ) is optimal, then


1

|Et | dt ⊂⊂ Ω.

spt 0

Indeed, we can find Ω  ⊂⊂ Ω which still has the property that the open r-neighbourhood of X is contained in Ω  , with r = diam(X), hence





1

1

|Et |(Ω) dt = min (MK) = 0

|Et |(Ω  ) dt.

0

(2) Since the two sides in the chain of inequalities (24) are equal when γ is optimal, we infer |πt# ((y − x)γ)| = πt# (|y − x|γ) Analogously, we have


  

1 0


 Et dt =

for L1 -a.e. t ∈ (0, 1).

(28)

1

|Et | dt

(29)

0

whenever (ft , Et ) is optimal. If the strict inequality < holds, then
fˆt = f0 + t(f1 − f0 )

ˆt = and E

1

Eτ dτ 0

provide an admissible pair for (ODE) with strictly less energy. Remark 4.3. [Nonlinear cost] When the cost function c(x, y) is |x − y|p for some p > 1 the corresponding problem (MK) is still equivalent to (ODE), provided we minimize, instead of J, the energy
1 Φp (ft , Et ) dt, Jp (f, E) := 0

Lecture Notes on Optimal Transport Problems

where


 p ν      σ  dσ X Φp (σ, ν) =    +∞

23

if |ν| << σ; otherwise.

The proof of the equivalence is quite similar to the one given in Theorem 4.2. Again the essential ingredient is the inequality Φp (ν ∗ ρε , σ ∗ ρε ) ≤ Φp (ν, σ). The latter follows by Jensen’s inequality and the convexity of the map (z, t) → |z|p t1−p in Rn × (0, ∞), which provide the pointwise estimate    p   ν ∗ ρ ε p  σ ∗ ρε ≤  ν  σ ∗ ρε .   σ ∗ ρε  σ In this case, under the same assumptions on f0 made in Corollary 2.2, due to the uniqueness of the optimal planning γ = (Id × ψ# )f0 , we obtain that, for optimal (ft , Et ), γ ε = (Id×ψ1ε )# f0ε converge to γ as ε → 0+ . As a byproduct, the maps ψ1ε converge to ψ as Young measures. If f0 << Ln it follows that ψh converge to ψ in [Lr (f0 )]n for any r ∈ [1, ∞) (see Lemma 9.1 and the remark following it). See §2.6 of [4] for a systematic analysis of the continuity and semicontinuity properties of the functional (σ, ν) → Φp (σ, ν) with respect to weak convergence of measures. I do not know whether in the previous proof it is actually possible to show full convergence of γ ε as ε → 0+ . However, using a more refined estimate and a geometric lemma (see Lemma 4.1 below) we will prove that any limit measure γ satisfies the condition
1
1 πt# (|y − x|γ) dt = |Et | dt. (30) 0

0

We call transport density any of the measures in (30). This double representation will be relevant in Section 7, where under suitable assumptions we will obtain the uniqueness of the transport density. Corollary 4.1. Let (ft , Et ) be optimal for (ODE). Then there exists an op1 timal planning γ such that (30) holds. In particular spt 0 |Et | dt ⊂ X. Proof. Let m = min (MK) = min (ODE) and recall that, by Remark 4.2(1), 1 the measure 0 |Et | dt has compact support in Ω. It suffices to build an optimal planning γ such that
1
1 πt# ((y − x)γ) dt = Et dt. (31) 0

0

24

Luigi Ambrosio

Indeed, by (29) we get





1

πt# (|y − x|γ) ≥ 0

1

|Et | dt 0

and the two measures coincide, having both mass equal to m. Keeping the same notation used in the final part of the proof of Theorem 4.2, we define γtε as (Id × ψtε )# f0ε and we compute


1

I(γ ε ) = 0

d I(γtε ) dt = dt




1


BR

0

ψtε (x) − x ˙ ε · ψ (x)f0ε (x) dxdt. |ψtε (x) − x| t

Since I(γ ε ) → m as ε → 0+ and since (by (25))


1







1




|ψ˙ tε (x)|f0ε (x) dxdt = BR

0


|Etε | dxdt

0

BR

≤

1

|Et |(Ω) dt = m, 0

we infer





1

lim

ε→0+

BR

0

2   ψ˙ ε (x) ψtε (x) − x  ˙ ε  t −  |ψt (x)|f0ε (x) dtdx = 0.  ε  |ψ˙ t (x)| |ψtε (x) − x| 

(32)

Now, using Lemma 4.1 and the Young inequality 2ab ≤ δa2 + b2 /δ, for any φ ∈ Cc∞ (Rn ) we obtain 
1     [πt# ((y − x)γ ε ) − Etε ](φ) dt   0 

1    ε ε ε ε ε ˙  = [(ψ1 (x) − x)φ(t(ψ1 (x) − x)) − ψt (x)φ(ψt (x))]f0 (x) dtdx X 0
≤ R Lip(φ)δ Length(ψtε )f0ε (x) dx X

R Lip(φ) + δ



X

0

1

2   ψ˙ ε (x) ψtε (x) − x  ˙ ε  t −  |ψt (x)|f0ε (x) dtdx.  ε  |ψ˙ t (x)| |ψtε (x) − x| 

with R = diam(Ω). Passing to the limit first as ε → 0+ , taking (32) into account, and then passing to the limit as δ → 0+ , we obtain
1
1 πt# ((y − x)γ)(φ) dt = Et (φ) dt. 0

0

Remark 4.4. [Decomposition of vectorfields] Let E be a vectorfield with ∇ · E = f in Ω and assume that |E|(Ω) = min (MK) with f0 = f + and f1 = f − . Then, choosing Et = E and ft = f0 + t(f1 − f0 ) in (31), we obtain that E can be represented as the superposition of elementary vectorfields

Lecture Notes on Optimal Transport Problems

25


E=

Exy dγ(x, y) Ω×Ω

with

Exy := (y − x) H1 [ x, y]]

without cancellations, i.e. in such a way that

|Exy | dγ(x, y), |∇ · E| = |E| = Ω×Ω

|∇ · Ex,y | dγ(x, y).

Ω×Ω

The possibility to represent a generic vectorfield whose divergence is a measure as a superposition without cancellations of elementary vectorfields associated to rectifiable curves is discussed in [39]. It turns out that this property is not true in general, at least if n ≥ 3; hence the optimality condition |E|(Ω) = min (MK) is essential. Lemma 4.1. Let ψ ∈ Lip([0, 1], Rn ) with ψ(0) = 0 and φ ∈ Lip(Rn ). Then  
1 
1  ˙    ψ(t) ψ(t)  ˙  ˙   ≤ LR − ψ(1)φ(tψ(1)) − ψ(t)φ(ψ(t)) dt  |ψ(t)| dt,    ˙   |ψ(t)| | ψ(t)| 0 0 with L = Lip(φ), R = sup |ψ|. Proof. We start from the elementary identity  d d [φ(sψ(t))ψi (t)] − sφ(sψ(t))ψ˙ i (t) dt ds n  

∂φ =s (sψ(t)) ψi (t)ψ˙ j (t) − ψj (t)ψ˙ i (t) ∂xj j=1

(33)

and integrate both sides in [0, 1] × [0, 1]. Then, the left side becomes exactly the integral that we need to estimate from above. The right side, up to the multiplicative constant Lip(φ), can be estimated with 


1
1   ˙ ψ(t) ψ(t)   ˙ ˙ ∧ − |ψ(t) ∧ ψ(t)| dt = |ψ(t)| ψ(t)  dt ˙   |ψ(t)| |ψ(t)| 0 0  
1 ˙ ψ(t)  ˙  ψ(t) − ≤ sup |ψ|  |ψ(t)| dt.  ˙ |ψ(t)|  0  |ψ(t)| Remark 4.5. The geometric meaning of the proof above is the following: the integral to be estimated is the action on the 1-form φdxi of the closed and rectifiable current T associated to the closed path starting at 0, arriving at ψ(1) following the curve ψ(t), and then going back to 0 through the segment [ 0, ψ(1)]]; for any 2-dimensional current G such that ∂G = T , as T (φdxi ) = G(dφ ∧ dxi ),

(34)

26

Luigi Ambrosio

this action can be estimated by the mass of G times Lip(φ). The cone construction provides a current G with ∂G = T , whose mass can be estimated by
1
1 ˙ s|ψ(t) ∧ ψ(t)| dsdt. 0

0

With this choice of G the identity (34) corresponds to (33). In order to relate also (ft , Et ) to the Kantorovich potential, we define the transport rays and the transport set and we prove the differentiability of the potential on the transport set. Definition 4.1 (Transport rays and transport set). Let u ∈ Lip1 (X). We say that a segment ] x, y[[⊂ X is a transport ray if it is a maximal open oriented segment whose endpoints x, y satisfy the condition u(x) − u(y) = |x − y|.

(35)

The transport set Tu is defined as the union of all transport rays. We also define Tue as the union of the closures of all transport rays. Denoting by F the compact collection of pairs (x, y) with x = y such that (35) holds, the transport set is also given by   {x + t(y − x)} Tu = t∈(0,1) (x,y)∈F

and therefore is a Borel set (precisely a countable union of closed sets). We can now easily prove that the u is differentiable at any point in Tu . Proposition 4.2 (Differentiability of the potential). Let u ∈ Lip1 (X). Then u is differentiable at any point z ∈ Tu . Moreover −∇u(z) is the unit vector parallel to the transport ray containing z. Proof. Let x, y ∈ X be such that (35) holds. By the triangle inequality and the 1-Lipschitz continuity of u we get u(x) − u(x + t(y − x)) = t|y − x|

∀t ∈ [0, 1].

This implies that, setting ν = (y − x)/|y − x|, the partial derivative of u along ν is equal to −1 for any internal point z of the segment. For any unit vector ξ perpendicular to ν we have   u(z + hξ) − u(z) = u(z + hξ) − u(z + |h|ν) + u(z + |h|ν) − u(z)   ≤ |h|2 + |h| − |h| = O(|h|3/2 ) = o(|h|). A similar argument also proves that u(z + hξ) − u(z) ≥ o(|h|). This proves the differentiability of u at z and the identity ∇u(z) = −ν.

Lecture Notes on Optimal Transport Problems

27

In Section 6 we will need a mild Lipschitz property of the potential ∇u on Tue (see also [15, 43] for the case of general strictly convex norms). We will use this property to prove in Corollary 6.1 that Tue \ Tu is Lebesgue negligible. This property can also be used to prove that ∇u is approximately differentiable Ln -a.e. on Tu (see for instance Theorem 3.1.9 of [24]). Theorem 4.3 (Countable Lipschitz property). Let u ∈ Lip1 (X). There exists a sequence of Borel sets Th covering Ln -almost all of Tue such that ∇u restricted to Th is a Lipschitz function. Proof. Given a direction ν ∈ Sn−1 and a ∈ R, let R be the union of the half closed transport rays [ x, y[[ with y − x, ν ≥ 0 and y, ν ≥ a. It suffices to prove that the restriction of ∇u to T := R ∩ {x : x · ν < a} \ Σu has the countable Lipschitz property stated in the theorem. To this aim, since BVloc funtions have this property (see for instance Theorem 5.34 of [4] or [24]), it suffices to prove that ∇u coincides Ln -a.e. in T with a suitable function w ∈ [BVloc (Sa )]n , where Sa = {x : x · ν < a}. To this aim we define u ˜(x) := min {u(y) + |x − y| : y ∈ Ya } , where Ya is the collection of all right endpoints of transport rays with y·ν ≥ a. By construction u ˜ ≥ u and equality holds on R ⊃ T . We claim that, for b < a, u ˜ − C|x|2 is concave in Sb for C = C(b) large enough. Indeed, since |x − y| ≥ a − b > 0 for any y ∈ Ya and any x ∈ Sb , the functions u(y) + |x − y| − C|x|2 , y ∈ Ya are all concave in H for C large enough depending on a − b. In particular, as gradients of real valued concave functions are BVloc (see for instance [2]), we obtain that w := ∇˜ u = ∇(u − C|x|2 ) + 2Cx is a BVloc function in Sa . Since ∇u = w Ln -a.e. in T the proof is achieved. By a similar argument, using semi-convexity in place of semi-concavity, one can take into account the right extreme points of the transport rays. As a byproduct of the equivalence between (MK) and (ODE) we can prove that (ODE) has “regular” solutions, related to the transport set and to the gradient of any maximal Kantorovich potential u. Theorem 4.4 (Regularity of (ft , Et )). For any solution (ft , Et ) of (ODE) representable as in Theorem 4.2 for a suitable optimal planning γ, there exists a 1-Lipschitz function u such that (i) ft is concentrated on the transport set Tu and |Et | ≤ Cft for L1 -a.e. t ∈ (0, 1), with C = diam(X);

28

Luigi Ambrosio

(ii) Et = −∇u|Et | for L1 -a.e. t ∈ (0, 1); (iii) for any convolution kernel ρ we have
1
lim |∇u − ∇u ∗ ρε |2 d|Et | dt = 0; (iv) |Et |(X) =

1 0

ε→0+

0

(36)

X

|Eτ |(X) dτ for L1 -a.e. t ∈ (0, 1).

Proof. (i) Let u be any maximal Kantorovich potential and let T = Tu be the associated transport set. For any t ∈ (0, 1) we have
χ{(x,y): x+t(y−x)∈T ft (X \ T ) = / } dγ = 0 X×X

because, by (12), any segment ] x, y[[ with (x, y) ∈ spt γ is contained in a transport ray, hence contained in T . The inequality |Et | ≤ Cft simply follows by the fact that |x − y| ≤ C on spt γ. (ii) Choosing t satisfying (28) and taking into account Proposition 4.2, for any φ ∈ C(X) we obtain

φ∇u d|Et | = φ(πt )∇u(πt )|y − x| dγ X X×X
φ(πt )(y − x) dγ = −Et (φ). =− X×X

(iii) Let m = J(E) = min (MK). By (13) we get

u d(f0 − f1 ) = lim u ∗ ρε d(f1 − f0 ) m= X








X




1

u ∗ ρε df˙t dt = lim

= lim

ε→0+

ε→0+

X

ε→0+

0

1

∇u ∗ ρε · Et dt 0

1

≤

|Et |(X) dt = m. 0

This proves that lim+

ε→0




1




|Et |(X) −

0

 ∇u ∗ ρε , ∇u d|Et | dt = 0,

X

whence, taking into account that |∇u| = 1 and |∇u ∗ ρε | ≤ 1, (36) follows. (iv) The inequality |Et |(X) ≤ I(γ) = J(E) has been estabilished during the proof of Theorem 4.2. By minimality equality, it holds for L1 -a.e. t ∈ (0, 1). Remark 4.6. (1) It is easy to produce examples of optimal flows (ft , Et ) satisfying (i), (ii), (iii), (iv) which are not representable as in Theorem 4.2 (again it suffices to consider the sum of two paths, with constant sum of velocities). It is not clear which conditions must be added in order to obtain this representation property. (2) Notice that condition (i) actually implies that f is a Lipschitz map from [0, 1] into M1 (X) endowed with the 1-Wasserstein metric.

Lecture Notes on Optimal Transport Problems

29

5 The PDE version of the optimal transport problem and the p-laplacian approximation In this section we see how, in the case when the cost function is linear, the optimal transport problem can be rephrased using a PDE. As in the previous section we consider an auxiliary bounded open set Ω such that the open rneighbourhood of X is contained in Ω, with r > diam(X); we assume also that Ω has a Lipschitz boundary. Specifically, we are going to consider the following problem and its connections with (MK) and (ODE). (PDE) Let f ∈ M(X) be a measure with f (X) = 0. Find µ ∈ M+ (Ω) and u ∈ Lip1 (Ω) such that (i) there exist smooth functions uh uniformly converging to u on X, equal to 0 on ∂Ω and such that the functions ∇uh converge in [L2 (µ)]n to a function ∇µ u satisfying |∇µ u| = 1

µ-a.e. in Ω;

(37)

(ii) the following PDE is satisfied in the sense of distributions −∇ · (∇µ u µ) = f

in Ω.

(38)

Given (µ, u) admissible for (PDE), we call µ the transport density (the reason for that soon will be clear) and ∇µ u the tangential gradient of u. Remark 5.1. (1) Choosing uh as test function in (38) gives

2 |∇µ u| dµ = lim ∇µ u, ∇uh  dµ µ(Ω) = h→∞

Ω

(39)

Ω

= lim f (uh ) = f (u), h→∞

so that the total mass µ(Ω) depends only on f and u. We will prove in Theorem 5.1 that actually µ(Ω) depends only on f and that µ is concentrated on X. (2) It is easy to prove that, given (µ, u), there is at most one tangential ˜ µ u, we have uh → ∇ gradient ∇µ u: indeed, if ∇uh → ∇µ u and ∇˜

˜ µ u dµ = lim ∇µ u · ∇ ∇µ u · ∇˜ uh dµ = lim f (˜ uh ) = f (u) = µ(Ω), Ω

h→∞

Ω

h→∞

˜ µ u µ-a.e. On the other hand, Example 5.1 below shows that whence ∇µ u = ∇ to some u there could correspond more than one measures µ solving (PDE). We will obtain uniqueness in Section 7 under absolute continuity assumptions on f + or f − .

30

Luigi Ambrosio

We will need the following lemma, in which the assumption that Ω is “large enough” plays a role. Lemma 5.1. If (µ, u) solves (PDE) then spt µ ⊂⊂ Ω. Proof. Let Ω  ⊂⊂ Ω  ⊂⊂ Ω be such that |x − y| > r = diam(X) for any x ∈ X and any y ∈ Rn \Ω  . We choose a minimum point x0 for the restriction of u to X and define   w(x) := min [u(x) − u(x0 )]+ , dist(x, Rn \ Ω  ) . Then, it is easy to check that w(x) = u(x) − u(x0 ) on X and that w ≡ 0 on Rn \ Ω  . Since µ(Ω) = f (u) = f (u − u(x0 )) = lim f (w ∗ ρε )
→0+
= lim ∇µ u · ∇w ∗ ρε dx ≤ µ(Ω  ),
→0+

Ω

we conclude that µ(Ω \ Ω  ) = 0. In the following theorem we build a solution of (PDE) choosing µ as a transport density of (MK) with f0 = f + and f1 = f − . As a byproduct, by Corollary 4.1 we obtain that the support of µ is contained in X and that µ is representable as the right side in (30). Theorem 5.1 ((ODE) versus (PDE)). Assume that (ft , Et ) with


1

|Et | dt ⊂⊂ Ω

spt 0

solves (ODE) and let u be a maximal Kantorovich potential relative to f0 , f1 . 1 Then, setting f = f0 − f1 , the pair (µ, u), with µ = 0 |Et | dt, solves (PDE) with ∇µ u = ∇u and, in particular, spt µ ⊂ X. Conversely, if (µ, u) solves (PDE), setting f0 = f + and f1 = f − , the measures (ft , Et ) defined by ft := f0 + t(f1 − f0 ),

Et := −∇µ u µ

solve (ODE) and spt µ ⊂ X. In particular µ(X) = min (ODE) = min (MK). Proof. By Corollary 4.1 we can assume that µ is representable as in (30) for a suitable optimal planning γ. Then, condition (i) in (PDE) with ∇µ u = ∇u follows by (36). By (17) and condition (ii) of Theorem 4.4 we get ∇ · (∇µ u|Ft |) = ∇ · (∇u|Ft |) = f˙t ,

Lecture Notes on Optimal Transport Problems

31

with Ft = πt# ((y − x)γ). Integrating in time and taking into account the definition of µ, we obtain (38). Notice also that µ(X) = J(E). If (ft , Et ) are defined as above, we get f˙t + ∇ · Et = f1 − f0 − ∇ · (∇µ uµ) = f1 − f0 + f = 0. Also in this case J(E) = µ(X). Example 5.1 (Non uniqueness of µ). In general, given f , the measure which solves (PDE) for a given u ∈ Lip1 (X) is not unique. As an example one can consider the situation in Example 1.5, where µ can be concentrated either on the horizontal sides of the square or on the vertical sides of the square. As shown in [22], an alternative construction can be obtained by solving the problem “−∆∞ u = f ”, i.e. studying the following problems   −∇ · |∇up |p−2 ∇up = f in Ω , (40) up = 0 on ∂Ω as p → ∞. In this case µ is the limit, up to subsequences, of |∇up |p−1 Ln . Under suitable regularity assumptions, Evans and Gangbo prove that µ = aLn for some a ∈ L1 (Ω) and use (a, u) to build an optimal transport ψ; their construction is based on a careful regularization corresponding to the one used in Theorem 4.2 in the special case Et = aLn and ft = f0 + t(f1 − f0 ). However, since the goal in Theorem 4.2 is to build only an optimal planning and not an optimal transport, our proof is much simpler and works in a much greater generality (i.e., with no special assumption on f ). Now we also prove that the limiting procedure of Evans and Gangbo leads to solutions of (PDE), regardless of any assumption on the data f0 , f1 . Theorem 5.2. Let up be solutions of (40) with p ≥ n + 1. Then (i) the measures Hp = |∇up |p−2 ∇up Ln Ω are equi-bounded in Ω; (ii) (up ) are equibounded and equicontinuous in Ω; (iii) If (H, u) = limj (Hpj , upj ) with pj → ∞, then (µ, u) solves (PDE) with µ = |H|. Proof. (i) Using up as test function and the Sobolev embedding theorem, we get 



|∇up |p dx = f (up ) ≤ |f |(X)up ∞ ≤ C Ω

Ω

older inequality we infer with p0 = n + 1. Using H¨
p−p0 p |∇up |p dx ≤ C p−1 Ln (Ω) p0 (p−1) . Ω

1/p0

|∇up0 |p0 dx

,

32

Luigi Ambrosio

(ii) Follows by (i) and the Sobolev embedding theorem. (iii) Clearly −∇ · H = f in Ω. By (i) and H¨ older inequality we infer
|∇u|q dx < ∞, sup q≥n+1

Ω

whence |∇u| ≤ 1 L -a.e. in Ω. We notice first that the functional 2
  ν   ν →  |ν| − w φ d|ν| Ω n

is lower semicontinuous with respect to the weak convergence of measures for any nonnegative φ ∈ Cc (Ω), w ∈ [C(Ω)]n . The verification of this fact is straightforward: it suffices to expand the squares. Second, we notice that 2
  Hp   lim lim sup − ∇uε  d|Hp | = 0  + |H |
→0 p→∞ p Ω whenever uε ∈ Lip1 (Rn ) are smooth functions uniformly converging to u in X and equal to 0 on ∂Ω. Indeed, as |∇uε | ≤ 1, we have 2  

 Hp  ∇uε · ∇up p−1   d|Hp | ≤ 2 − ∇u dx 1 − |∇u | ε p   |∇up | Ω |Hp |
Ω ≤2 |∇up |p−2 (|∇up |2 − ∇uε · ∇up ) dx + ωp Ω

= 2f (up ) − 2f (uε ) + ωp , where ωp = supt≥0 tp−1 − tp tends to 0 as p → ∞. Taking into account these two remarks, setting p = pj and passing to the limit as j → ∞, we obtain 2
 H    φ d|H| = 0. lim − ∇u ε  
→0+ Ω |H| for any nonnegative φ ∈ Cc (Ω). This implies that ∇uε converge in L2loc (|H|) to the Radon–Nikod´ ym derivative H/|H|. Since |∇uε | ≤ 1 we have also convergence in [L2 (|H|)]n . We now set µ = |H| and ∇µ u = lim
∇u
, so that H = ∇µ uµ and (38) is satisfied.

6 Existence of optimal transport maps In this section we prove the existence of optimal transport maps in the case when c(x, y) = |x − y| and f0 is absolutely continuous with respect to Ln ,

Lecture Notes on Optimal Transport Problems

33

following essentially the original Sudakov approach and filling a gap in his original proof (see the comments after Theorem 6.1). Using a maximal Kantorovich potential we decompose almost all of X in transport rays and we build an optimal transport maps by gluing the 1-dimensional transport maps obtained in each ray. The assumption f0 << Ln is used to prove that the conditional measures f0C within any transport ray C are non atomic (and even absolutely continuous with respect to H1 C), so that Theorem 3.1 is applicable. Therefore the proof depends on the following two results whose proof is based on the countable Lipschitz property of ∇u stated in Theorem 4.3. Theorem 6.1. Let B ∈ B(X) and let π : B → Sc (X) be a Borel map satisfying the conditions (i) π(x) ∩ π(x ) = ∅ whenever π(x) = π(x ); (ii) x ∈ π(x) for any x ∈ B; (iii) the direction ν(x) of π(x) is a Sn−1 -valued countably Lipschitz map on B. Then, for any measure λ ∈ M+ (X) absolutely continuous with respect to Ln B, setting µ = π# λ ∈ M+ (Sc (X)), the measures λC of Theorem 9.1 are absolutely continuous with respect to H1 C for µ-a.e. C ∈ Sc (X). Proof. Being the property stated stable under countable disjoint unions we may assume that (a) there exists a unit vector ν such that ν(x) · ν ≥ (b) ν(x) is a Lipschitz map on B; (c) B is contained in a strip

1 2

for any x ∈ B;

{x : a − b ≤ x · ν ≤ a} with b > 0 sufficiently small (depending only on the Lipschitz constant of ν) and π(x) intersects the hyperplane {x : x · ν = a}. Assuming with no loss of generality ν = en and a = 0, we write x = (y, z) with y ∈ Rn−1 and z < 0. Under assumption (a), the map T : π(B) → Rn−1 which associates to any segment π(x) the vector y ∈ Rn−1 such that (y, 0) ∈ π(x) is well defined. Moreover, by condition (i), T is one to one. Hence, setting f = T ◦ π : B → Rn−1 , ν = T# µ = f# λ,

C(y) = T −1 (y) ⊃ f −1 (y)

and representing λ = ηy ⊗ ν with ηy = λC(y) ∈ M1 (f −1 (y)), we need only to prove that ηy << H1 C(y) for ν-a.e. y. To this aim we examine the Jacobian, in the y variables, of the map f (y, t). Writing ν = (νy , νt ), we have f (y, t) = y + τ (y, t)νy (y, t)

with

τ (y, t) = −

t . νt (y, t)

34

Luigi Ambrosio

Since νt ≥ 1/2 and τ ≤ 2b on B, we have   t det (∇y f (y, t)) = det Id + τ ∇y νy + 2 ∇y νt ⊗ νy > 0 νt if b is small enough, depending only on Lip(ν). Therefore, the coarea factor 

det2 A Cf := A

(where the sum runs on all (n − 1) × (n − 1) minors A of ∇f ) of f is strictly positive on B and, writing λ = gLn with g = 0 out of B, Federer’s coarea formula (see for instance [4], [38]) gives λ=

g g 1 Cf Ln = H Cf Cf

f −1 (y) ⊗ Ln−1 = ηy ⊗ ν  ,

  with L = y ∈ Rn−1 : H1 (f −1 (y)) > 0 and


g 1 f −1 (y) g Cf H   1 d H Ln−1 ηy :=  , ν = 1 Cf g/Cf d −1 H −1 f (y) f (y)

L.

By Theorem 9.2 we obtain ν = ν  and ηy = ηy for ν-a.e. y, and this concludes the proof. Remark 6.1. (1) In [42] V.N.Sudakov stated the theorem above (see Proposition 78 therein) for maps π : B → So (X) without the countable Lipschitz assumption (iii) and also in a greater generality, i.e. for a generic “Borel decomposition” of the space in open affine regions, even of different dimensions. However, it turns out that the assumption (iii) is essential even if we restrict to 1-dimensional decompositions. Indeed, G.Alberti, B.Kirchheim and D.Preiss [3] have recently found an example of a compact family of open and pairwise disjoint segments in R3 such that the collection B of the midpoints of the segments has strictly positive Lebesgue measure. In this case, of course, if λ = Ln B the conditional measures λC are unit Dirac masses concentrated at the midpoint of C, so that the conclusion of Theorem 6.1 fails. If n = 2 it is not hard to prove that actually condition (iii) follows by (ii), so that the counterexample mentioned above is optimal. (2) Since any open segment can be approximated from inside by closed segments, by a simple approximation argument one can prove that Theorem 6.1 still holds as well for maps π : B → So (X) or for maps with values into half open [ x, y[[ segments. Also the assumption (i) (that the segments do not intersect) can be relaxed. For instance we can assume that the segments can intersect only

Lecture Notes on Optimal Transport Problems

35

at their extreme points, provided the collection of these extreme points is Lebesgue negligible. This is precisely the content of the next corollary; again, a counterexample in [28] shows that this property need not be true for a generic compact family of disjoint line segments in Rn , n ≥ 3. Corollary 6.1 (Negligible extreme points). Let u ∈ Lip1 (X). Then the collection of the extreme points of the transport rays is Lebesgue negligible. Proof. We prove that the collection L of all left extreme points is negligible, the proof for the right ones being similar. Let B = L \ Σu and set π(x) := [ x, x −

r(x) ∇u(x)]], 2

where r(x) is the length of the transport ray emanating from x (this ray is unique due to the differentiability of u at x). By Theorem 4.3 the map π has the countable Lipschitz property on B and, by construction, π(x) ∩ π(x ) = ∅ whenever x = x . By Theorem 6.1 we obtain
λC dν, λ= Sc (X)

with λ = Ln B, ν = π# λ and λC << H1 C probability measures concentrated on π −1 (C) for ν-a.e. C. Since π −1 (C) contains only one point for any C ∈ π(B), we obtain λC = 0 for ν-a.e. C, whence λ(B) = ν(Sc (X)) = 0. Theorem 6.2 (Sudakov). Let f0 , f1 ∈ M1 (X) and assume that f0 << Ln . Then there exists an optimal transport ψ mapping f0 to f1 . Moreover, if −1 f1 << Ln we can choose ψ so that ψ −1 is well defined f1 -a.e. and ψ# f1 = f0 . Proof. Let γ be an optimal planning. In the first two steps we assume that for f0 -a.e. x there exists y = x such that (x, y) ∈ spt γ.

(41)

This condition holds for instance if f0 ∧ f1 = 0 because in this case any optimal planning γ does not charge the diagonal ∆ of X × X, i.e. γ(∆) = 0 (otherwise h = π0# (χ∆ γ) = π1# (χ∆ γ) would be a nonzero measure less than f0 and f1 ). Let u ∈ Lip1 (X) be a maximal Kantorovich potential given by Corollary 2.1, i.e. a function satisfying u(x) − u(y) = |x − y|

γ-a.e. in X × X

(42)

for any optimal planning γ, and let Tu be the transport set relative to u (see Definition 4.1). In the following we set  = {x ∈ X \ Σu : ∃ y = x s.t. (x, y) ∈ spt γ} . X

36

Luigi Ambrosio

By (41) and the absolute continuity assumption we know that f0 is concen Notice also that for any x ∈ X  there exists a unique closed trated on X.  is a transport ray containing x: this follows by the fact that any x ∈ X differentiability point of u and by Proposition 4.2. Step 1. We define r : X × X → Sc (X) as the map which associates to any pair (x, y) the closed transport ray containing [ x, y]]. By (42) the map r is well defined γ-a.e. out of the diagonal ∆; moreover, being f0 concentrated  r is also well defined γ-a.e. on ∆. Hence, according to Theorem 9.1 we on X, can represent γ = γC ⊗ ν with ν := r# γ and (42) gives u(x) − u(y) = |x − y|

γC -a.e. in X × X,

(43)

for ν-a.e. C ∈ Sc (X). By the sufficiency part in Corollary 2.1, we infer that γC is an optimal planning relative to the probability measures f0C := π0# γC ,

f1C := π1# γC

for ν-a.e. C ∈ Sc (X). By (56) we infer
πt# γ(B) = πt# γC (B) dν(C) Sc (X)

Notice also that

(44)


F1 (f0 , f1 ) =

Indeed,
I(γ) =

∀t ∈ [0, 1], B ∈ B(X).

Sc (X)





|x − y| dγ =

Sc (X)

X×X

F1 (f0C , f1C ) dν(C).

(45)


|x − y| dγC dν(C) =

X×X

Sc (X)

I(γC ) dν(C)

and we know by (43) that γC is optimal for ν-a.e. C ∈ Sc (X).  → Sc (X) the natural map, so that π(x) is the Step 2. We denote by π : X  × X we obtain closed transport ray containing x. Since r = π ◦ π0 on X ν = r# γ = π# (π0# γ) = π# f0 . Moreover (44) gives
f0 (B) =

Sc (X)

f0C (B) dν(C)

∀B ∈ B(X).

(46)

 can intersect only at their right Notice that the segments π(x), x ∈ X, extreme point and that, by Corollary 6.1, the collection of these extreme

Lecture Notes on Optimal Transport Problems

37

points is Lebesgue negligible. As a consequence of (46) with ν = π# f0 and Remark 6.1(2), the measures f0C are absolutely continuous with respect to H1 C for µ-a.e. C ∈ Sc (X). Hence, by Theorem 3.1, for µ-a.e. C ∈ Sc (X) we can find a nondecreasing map ψC : C → C (this notion makes sense, since C is oriented) such that ψC# f0C = f1C . Notice also that ψ −1 is well defined f1C -a.e., if also f1C << H1 C.  are pairwise Taking into account that the closed transport rays in π(X)  we can glue all the maps ψC to produce a single Borel map disjoint in X,  → X. The map ψ is Borel because we have been able to exhibit the one ψ:X dimensional transport map constructively and because of the Borel property of the maps C → fiC (see (15) and (54)); the simple but boring details are left to the reader. Since ψ# f0C = ψC# f0C = f1C for µ-a.e. C ∈ Sc (X), taking (44) into account we infer

ψ# f0C dν(C) = f1C dν(C) = f1 . ψ# f0 = Sc (X)

Sc (X)

Finally ψ is an optimal transport because (45) holds and any ψC is an optimal transport. Step 3. In this step we show how the assumption (41) can be removed. We define X  := {x ∈ X : (x, x) ∈ spt γ and (x, y) ∈ / spt γ ∀y = x} and L = {(x, x) : x ∈ X  }. Then, we set f0 = f0 with X  = X \ X  , and f1 := π1# (γ

L),

X  and f0 = f0

X  ,

f1 := f1 − f1 .

Since L ⊂ ∆, we have f1 = π0# (γ L) = f0 , hence we can choose on X  the map ψ1 = Id as transport map to obtain (ψ1 )# f0 = f1 . Since f0 is concentrated on X  , the condition (41) is satisfied with f0 in place of f0 and we can find an optimal transport map ψ2 : X  → X such that (ψ2 )# f0 = f1 . Gluing these two transport maps we obtain a transport map ψ such that ψ# f0 = f1 ; since, by construction, u(x) − u(ψ(x)) = |x − ψ(x)|

f0 -a.e. in X,

we infer that ψ is optimal. This proof strongly depends on the strict convexity of the euclidean distance, which provides the first and second order differentiability properties of the potential u on the transport set Tu . Notice also that if c(x, y) = x − y and the norm  ·  is not strictly convex, then the “transport rays” need not be one-dimensional and, to our knowledge, the existence of an optimal transport map is an open problem in this situation. Indeed, this existence result

38

Luigi Ambrosio

is stated by Sudakov in [42] but his proof is faulty, for the reasons outlined in Remark 6.1(1). Under special assumptions on the data f0 , f1 (absolute continuity, separated supports, Lipschitz densities), Evans and Gangbo provided in [22] a proof based on differential methods of the existence of optimal transport maps. For stricly convex norms, the first fully rigorous proofs of the existence of an optimal transport map under the only assumption that f0 << Ln have been given in [15] and [43]. As in Theorem 6.2, the proof is strongly based on the differentiability of the directions of transport rays.

7 Regularity and uniqueness of the transport density In this section we investigate the regularity and the uniqueness properties of the transport density µ arising in (PDE). Recall that, as Theorem 5.1 1 shows, any such measure µ can also be represented as 0 |Et | dt for a suitable optimal pair (ft , Et ) for (ODE) (even with Et independent of t). In turn, by 1 Corollary 4.1, any optimal measure µ = 0 |Et | dt can be represented as


1

πt# (|y − x|γ) dt

µ=

(47)

0

for a suitable optimal planning γ. For this reason, in the following we restrict our attention to the representation (47), valid for any optimal measure µ for (PDE). A more manageable formula for µ, first considered by G.Bouchitt´e and G.Buttazzo in [14], is given in the following elementary lemma. Lemma 7.1. Let µ be as in (47). Then
µ(B) = Length (]]x, y[[∩B) dγ(x, y)

∀B ∈ B(X).

(48)

X×X

Proof. For any Borel set B we have
1

µ(B) = |y − x| dγ(x, y) dt = 0




πt−1 (B)




1

|y − x| X×X

0

χπ−1 (B) dt dγ(x, y) t

Length (]]x, y[[∩B) dγ(x, y).

= X×X

A direct consequence of (48) is that µ is concentrated on the transport set (since γ-a.e. segment ] x, y[[ is contained in a transport ray); another consequence is the density estimate µ(Br (x)) ≤ f0 (X)f1 (X), 2r

(49)

Lecture Notes on Optimal Transport Problems

39

because Length(]]x, y[[∩Br (x)) ≤ 2r for any ball Br (x) and any segment ] x, y[[. The first results on µ that we state, proved by A.Pratelli in [34] (see also [19]), relate the dimensions of f0 and f1 to the dimension of µ and show that necessarily µ is absolutely continuous with respect to the Lebesgue measure if f0 (or, by symmetry, f1 ) has this property. Theorem 7.1. Assume that for some k > 0 we have f0 (Br (x)) <∞ rk r∈(0,1) sup

f0 -a.e. in X.

Then µ has the same property. In particular µ << Ln if f0 << Ln . The proof of Theorem 7.1 is based on (48); the density estimate on µ is achieved by a careful analysis of the transport rays crossing a generic ball Br (x). A similar analysis proved the following summability estimate (see [19]). Theorem 7.2. Assume that f0 = g0 Ln and f1 = g1 Ln with g0 , g1 ∈ Lp (X), p > 1. Then µ = hLn , with h ∈ L∞ (X) if p = ∞ and h ∈ Lq (X) for any q < p if p < ∞. It is not known whether g0 , g1 ∈ Lp implies h ∈ Lp for p < ∞. Definition 7.1 (Hausdorff dimension of a measure). Let µ ∈ M+ (X). The Hausdorff dimension H-dim(µ) is the supremum of all k ≥ 0 such that µ << Hk . In other words µ(B) = 0 whenever Hk (B) = 0 for some k < H-dim(µ) and for any k > H-dim(µ) there exists a Borel set B with µ(B) > 0 and Hk (B) = 0. Notice that if µ is made of pieces of different dimensions, then H-dim(µ) is the smallest of these dimensions. Using the density estimates and the implications (see for instance Theorem 2.56 of [4] or [38]; here t > 0 and k > 0) µ(Br (x)) ≥ t ∀x ∈ B =⇒ µ(B) ≥ t Hk (B), ωk r k µ(Br (x)) ≤ t ∀x ∈ B =⇒ µ(B) ≤ 2k t Hk (B), lim sup k ω r + k r→0 lim sup

(50)

r→0+

we can prove a natural lower bound on the Hausdorff dimension of µ. Corollary 7.1. Let µ be a transport density. Then H-dim(µ) ≥ max {1, H-dim(f0 ), H-dim(f1 )} .

(51)

40

Luigi Ambrosio

Proof. By (49) we infer that µ has finite (and even bounded) 1-dimensional density at any point. In particular (51) gives µ(B) = 0 whenever H1 (B) = 0, so that H-dim(µ) ≥ 1. Let k = H-dim(f0 ) and k  < k; then f0 has finite k  -dimensional density
f0 -a.e., otherwise by (50) the set where the density is not finite would be Hk negligible and with strictly positive f0 -measure. By Theorem 7.1 we infer that µ has finite k  -dimensional density µ-a.e. By the same argument used before with k  = 1 we obtain that H-dim(µ) ≥ k  and therefore, since k  < k is arbitrary, H-dim(µ) ≥ k. A symmetric argument proves that H-dim(µ) ≥ H-dim(f1 ). We conclude this section proving the uniqueness of the transport density, under the assumption that either f0 << Ln or f1 << Ln (see Example 5.1 for a nonuniqueness example if neither f0 nor f1 are absolutely continuous). Similar results have been first announced by Feldman and McCann (see [25]). We first deal with the one dimensional case, where this absolute continuity assumption is not needed. Lemma 7.2. Let µ be a transport density in X ⊂ R and assume that the interior (a, b) of X is a transport ray. Then µ = hL1 in (a, b) with h(t) = σf ((a, t)) + c

L1 -a.e. in (a, b),

for some constant c, where f = f1 − f0 and σ = 1 if u = 1 in (a, b), σ = −1 if u = −1 in (a, b). Moreover  F1 (f0 , f1 ) − σ (a,b) (b − t) df (t) c= . b−a Proof. The proof of the first statement is based on the equation µ = σf and on a smoothing argument. By integrating both sides we get
c(b − a) = µ(X) − σ




b

b




f ((a, t)) dt = µ(X) − σ a







= µ(X) − σ

1 df (τ ) dt a

b




(a,t)

1 dt df (τ ) = µ(X) − σ (a,b)

τ

(b − τ ) df (τ ). (a,b)

The conclusion is achieved taking into account that µ(X) = min (MK) = F1 (f0 , f1 ). Theorem 7.3 (Uniqueness). Assume that either f0 or f1 are absolutely continuous with respect to Ln . Then the transport density is absolutely continuous and unique. Proof. We already know from Theorem 7.1 that any transport density is absolutely continuous and we can assume that f0 << Ln . By Corollary 4.1

Lecture Notes on Optimal Transport Problems

41

we know that the class of transport densities relative to (f0 , f1 ) is equal to the class of transport densities relative to (f0 − h, f1 − h) where h is any measure in M+ (X) such that h ≤ f0 ∧ f1 (indeed, this subtraction does not change the velocity field Et ). Hence, it is not restrictive to assume that f0 ∧ f1 = 0. We can assume that µ is representable as in (30) for a suitable optimal planning γ. Adopting the same notation of the proof of Theorem 6.1, we have γ = γC ⊗ ν, where ν = π# f0 does not depend on γ (and here the absolute continuity assumption on f0 plays a crucial role) and γC is an optimal planning relative to the measures f0C = π0# γC , f1C = π1# γC for ν-a.e. C. In particular
1
1 µ= πt# (|y − x|γC ⊗ ν) dt = πt# (|y − x|γC ) dt ⊗ ν. 0

0

Hence, it suffices to show that the measures
1 πt# (|y − x|γC ) dt µC := 0

do not depend on γ (up to ν-negligible sets of course). To this aim, taking into account Lemma 7.2, it suffices to show that f0C and f1C do not depend on γ. Indeed, we already know from (46) and Theorem 9.2 that f0C do not depend on γ. The argument for f1C is more involved. First, since γ(∆) = 0 (due to the assumption f0 ∧ f1 = 0), we have |x − y| > 0 γ-a.e., and therefore |x − y| > 0 γC -a.e. for ν-a.e. C ∈ Sc (X). As a consequence, setting C = [ xC , yC ] , we  obtain that f1C (xC ) = 0 for ν-a.e. C. Second, we examine the restriction f1C of f1C to the relative interior of C noticing that (44) gives

 f1 Tu = f1C Tu dν(C) = f1C dν(C), Sc (X)

Sc (X)

˜ By Theorem 9.2 because Tu ∩C is the relative interior of C for any C ∈ π(X).  we obtain that f1C depend only on f1 , Tu and ν. In conclusion, since   f1C = f1C + (1 − f1C (X)) δyC ,

we obtain that f1C do not depend on γ as well.

8 The Bouchitt´ e–Buttazzo mass optimization problem In [13, 14] Bouchitt´e and Buttazzo consider the following problem. Given f ∈ M(X) with f (X) = 0, they define

42

Luigi Ambrosio


E(µ) := inf X

1 |∇v|2 dµ − f (v) : v ∈ C ∞ (X) 2



for any µ ∈ M+ (X). Then, they raised the following mass optimization problem. (BB) Given m > 0, maximize E(µ) among all measures µ ∈ M+ (X) with µ(X) = m. A possible physical interpretation of this problem is the following: we may imagine that µ represents the conductivity of some material, thinking that the conductivity (i.e. the inverse resistivity) is zero out of spt µ; accordingly, we may imagine that f = f + − f − is a balanced density of positive and negative charges. Then −E(µ) represents the heating corresponding to the given conductivity, so that there is an obvious interest in maximizing E(µ) and, for a minimizer u, the (formal) first variation of the energy −∇ · (∇u µ) = f + − f − corresponds to Ohm’s law. More generally, problems of this sort appear in Shape Optimization and Linear Elasticity. In these cases u is no longer real valued and general energies of the form
F(u) = f (x, ∇u(x)) dµ(x) (52) X

must be considered in order to take into account light structures, corresponding to mass distributions not absolutely continuous with respect to Ln . This was the main motivation for Bouchitt´e, Buttazzo and Seppecher [12] in their development of a general differential calculus with measures. This calculus, based on a suitable concept of tangent space to a measure, enables the study of the relaxation of functionals (52) and provides an explicit formula for their lower semicontinuous envelope. Coming back to the scalar problem (BB), a remarkable fact discovered in [13, 14] is its connection with the Monge–Kantorovich optimal transport problem, in the case when c(x, y) = |x − y|. It turns out that the solutions of (BB) are in one to one correspondence with constant multiples of transport densities for (PDE) (or, equivalently, for (MK) with f0 = f + and f1 = f − ). As a byproduct we have that µ is unique and absolutely continuous with respect to Ln if either f0 or f1 are absolutely continuous with respect to Ln . Theorem 8.1 ((PDE) versus (BB)). Let (µ, u) be a solution of (PDE) and set m ˜ = µ(X). Then µ ˜= m m ˜ µ solves (BB) and any solution of (BB) is representable in this way. Moreover max(BB) = −

˜2 1m . 2 m

Lecture Notes on Optimal Transport Problems

43

Proof. Setting v = λuh (with uh as in the definition of (PDE)), for any ν ∈ M+ (X) with ν(X) = m we estimate E(ν) ≤

m 2 λ − λf (uh ), 2

so that, letting h → ∞ and taking into account (39), we obtain E(ν) ≤

m 2 λ − mλ. ˜ 2

By minimizing with respect to λ we obtain E(ν) ≤ −m ˜ 2 /(2m). On the other hand, using Young inequality and choosing λ so that λm ˜ = m, we can estimate

1 m |∇v|2 d˜ µ − f (v) ≥ ∇v, λ∇µ u d˜ µ − λ2 − f (v) 2 X 2 X ˜2 1m = −∇ · (∇µ u µ)(v) − f (v) − 2 m ˜2 1m , =− 2 m ˜ is optimal for (BB). for any v ∈ C ∞ (X). This proves that µ Conversely, it has been proved in [13, 14] that for any solution σ of (BB) there exists v ∈ Lip(X) such that |∇σ v| = m/m ˜ σ-a.e. and −∇ · (∇σ vσ) = f, where ∇σ v is understood in the Bouchitt´e–Buttazzo sense. But, since (see [14] again)
|∇σ v|2 dσ X  
= inf lim inf |∇vh |2 dσ : vh → v uniformly, vh ∈ C ∞ (X) , h→∞

X

we obtain a sequence of smooth functions vh uniformly converging to v such m ˜ that ∇vh → ∇σ v in [L2 (σ)]n . Setting u = m v and µ = m m ˜ σ, this proves that (σ, u) solves (PDE).

9 Appendix: some measure theoretic results In this section we list all the measure theoretic results used in the previous section; reference books for the content of this section are [37], [18] and [4]. Let us begin with some terminology. • (Measures) Let X be a locally compact and separable metric space. We denote by [M(X)]m the space of Radon measures with values in Rm and

44

Luigi Ambrosio

with finite total variation in X. We recall that the total variation measure of µ = (µ1 , . . . , µm ) ∈ [M(X)]m is defined by  ∞ ∞

 |µ(Bi )| : B = Bi , Bi ∈ B(X) |µ|(B) := sup i=1

i=1

and belongs to M+ (X). By Riesz theorem the space [M(X)]m , endowed with the norm µ = |µ|(X), is isometric to the dual of [C(X)]m . The duality is given by the integral, i.e. m


ui dµi . µ, u := i=1

X

Recall also that, for µ ∈ M+ (X) and f ∈ [L1 (X, µ)]m , the measure f µ ∈ [M(X)]m is defined by
f dµ ∀B ∈ B(X) f µ(B) := B

and |f µ| = |f |µ. • (Push forward of measures) Let µ ∈ [M(X)]m , let Y be another metric space and let f : X → Y be a Borel map. Then the push forward measure f# µ ∈ [M(Y )]m is defined by  ∀B ∈ B(Y ) f# µ(B) := µ f −1 (B) and satisfies the more general property

u df# µ = u ◦ f dµ for any bounded Borel function u : Y → R. Y

X

It is easy to check that |f# µ| ≤ f# |µ|. • (Support) We say that µ ∈ [M(X)]m is concentrated on a Borel set B if |µ|(X \ B) = 0 and we denote by spt µ the smallest closed set on which µ is concentrated (the existence of a smallest set follows by the separability of X, precisely by the Lindel¨ of property). The support is also given by the formula (sometimes taken as the definition) spt µ = {x ∈ X : |µ|(B (x)) > 0 ∀ > 0} . • (Convolution) If X ⊂ Rn , µ ∈ [M(X)]m and ρ ∈ Cc∞ (Rn ) (for instance a convolution kernel); we define
ρ(x − y) dµ(y) x ∈ Rn . µ ∗ ρ(x) := X

Notice that µ ∗ ρ ∈ [C ∞ (Rn )]m because Dα (µ ∗ ρ) = (Dα ρ) ∗ µ for any multiindex α and

Lecture Notes on Optimal Transport Problems

45

sup |Dα (µ ∗ ρ)| ≤ sup |Dα ρ|µ. Moreover, using Jensen’s inequality it is easy to check (see for instance Theorem 2.2(ii) of [4]) that
|µ ∗ ρ| dx ≤ |µ|(Cρ ), where Cρ := {x : dist(x, C) ≤ diam(spt ρ)} , C

(53) for any closed set C ⊂ Rn . • (Weak convergence) Assume that X is a compact metric space. We say that a family of measures (µh ) ⊂ M(X) weakly converges to µ ∈ M(X) if

lim u dµh = u dµ ∀u ∈ C(X), h→∞

X

X

i.e., if µh weakly∗ converge to µ as elements of the dual of C(X). In the case X is locally compact and separable, the concept is analogous, simply replacing C(X) the the subspace Cc (X) of functions with compact support. It is easy to check that µ ∗ ρε Ln weakly converge to µ in Rn whenever ρ is a convolution kernel. We mention also the following criterion for the weak convergence of positive measures (see for instance Proposition 1.80 of [4]): if µh , µ ∈ M+ (X) satisfy lim inf µh (A) ≥ µ(A) ∀A ⊂ X open h→∞

and lim sup µh (X) ≤ µ(X), 



h→∞

then X φ dµh → X φ dµ for any bounded continuous function φ : X → R. • (Measure valued maps) Let X, Y be locally compact and separable metric spaces. Let y → λy be a map which assigns to any y ∈ Y a measure λy ∈ [M(X)]m . We say that λy is a Borel map if y → λy (A) is a real valued Borel map for any open set A ⊂ X. By a monotone class argument it can be proved that y → λy is a Borel map if and only if y → λy ({x : (y, x) ∈ B})

is a Borel map for any B ∈ B(Y × X).

(54)

Moreover y → |λy | is a Borel map whenever y → λy is Borel (detailed proofs are in §2.5 of [4]). If µ ∈ M+ (Y ), analogous statements hold for B(Y )µ -measurable measure valued maps, where B(Y )µ is the σ-algebra of µmeasurable sets (in this case one has to replace B(Y × X) by B(Y )µ ⊗ B(X) in (54)). • (Decomposition of measures) The following result plays a fundamental role in these notes; it is also known as disintegration theorem.

46

Luigi Ambrosio

Theorem 9.1 (Decomposition of measures). Let X, Y be locally compact and separable metric spaces and let π : X → Y be a Borel map. Let λ ∈ [M(X)]m and set µ = π# |λ| ∈ M+ (Y ). Then there exist measures λy ∈ [M(X)]m such that (i) y → λy is a Borel map and |λy | is a probability measure in X for µ-a.e. y ∈Y; (ii) λ = λy ⊗ µ, i.e.
λ(A) = λy (A) dµ(y) ∀A ∈ B(X); (55) Y

 (iii) |λy | X \ π −1 (y) = 0 for µ-a.e. y ∈ Y . The representation provided by Theorem 9.1 of λ can be used sometimes to compute the push forward of λ. Indeed, f# (λy ⊗ µ) = f# λy ⊗ µ

(56)

for any Borel map f : X → Z, where Z is any other compact metric space. Notice also that |λ| = |λy | ⊗ µ. (57) Indeed, the inequality “≤” is trivial and the opposite one follows by evaluating both measures at B = X, using the fact that |λy |(X) = 1 for µ-a.e. y. In the case when m = 1 and λ ∈ M+ (X), the proof of Theorem 9.1 is available in many textbooks of measure theory or probability (in this case λy are the the so-called conditional probabilities induced by the random variable π, see for instance [18]); in the vector valued case one can argue component by component, but the fact that |λy | are probability measures is not straightforward. In Theorem 2.28 of [4] the decomposition theorem is proved in the case when X = Y × Z is a product space and π(y, z) = y is the projection on the first variable; in this situation, since λy are concentrated on π −1 (y) = {y}×Z, it is sometimes convenient to consider them as measures on Z, rather than measures on X, writing (55) in the form
λ(B) = λy ({z : (y, z) ∈ B}) dµ(y) ∀B ∈ B(X). (58) Y

Once the decomposition theorem is known in the special case X = Y × Z and π(y, z) = z the general case can be easily recovered: it suffices to embed X into the product Y × X through the map f (x) = (π(x), x) and to apply ˜ = f# λ. the decomposition theorem to λ Now we discuss the uniqueness of λx and µ in the representation λ = λx ⊗ µ. For simplicity we discuss only the case of positive measures.

Lecture Notes on Optimal Transport Problems

47

Theorem 9.2. Let X, Y and π be as in Theorem 9.1; let λ ∈ M+ (X), µ ∈ M+ (Y ) and let y → ηy be a Borel M+ (X)-valued map defined on Y such that  (i) λ = ηy ⊗ µ, i.e. λ(A) = Y ηy (A) dµ(y) for any A ∈ B(X); (ii) ηy (X \ π −1 (y)) = 0 for µ-a.e. y ∈ Y . Then ηy are uniquely determined µ-a.e. in Y by (i) and (ii), and, setting B = {y : ηy (X) > 0}, the measure µ B is absolutely continuous with respect to π# λ. In particular µ B ηy = λy π# λ

for π# λ-a.e. y ∈ Y ,

(59)

where λy are as in Theorem 9.1. Proof. Let ηy , ηy be satisfying (i) and (ii). We have to show that ηy = ηy for µ-a.e. y. Let (An ) be a sequence of open sets stable by finite intersection which generates the Borel σ-algebra of X. Choosing A = An ∩ π −1 (B), with B ∈ B(Y ), in (i) gives

ηy (An ) dµ(y) = ηy (An ) dµ(y). B

B

Being B arbitrary, we infer that ηy (An ) = ηy (An ) for µ-a.e. y, and therefore there exists a µ-negligible set N such that ηy (An ) = ηy (An ) for any n ∈ N and any y ∈ Y \ N . By Proposition 1.8 of [4] we obtain that ηy = ηy for any y ∈ Y \ N. Let B  ⊂ B be any π# λ-negligible set; then π −1 (B  ) is λ-negligible and therefore (ii) gives

 0= ηy π −1 (B  ) dµ(y) = ηy (X) dµ(y). Y

B


As ηy (X) > 0 on B ⊃ B  , this implies that µ(B  ) = 0. Writing µ B = hπ# λ, we obtain λ = hηy ⊗ π# λ and λ = λy ⊗ π# λ. As a consequence, (59) holds. • (Young measures) These measures, introduced by L.C.Young, arise in a natural way in the study of oscillatory phenomena and in the analysis of weak limits of nonlinear quantities (see [31] for a comprehensive introduction to this wide topic). Specifically, assume that we are given compact metric spaces X, Y and a sequence of Borel maps ψh : X → Y ; in order to understand the limit behaviour of ψh , we associate to them the measures
γψh = (Id × ψh )# µ = δψh (x) dµ(x)

48

Luigi Ambrosio

and we study their limit in M(X ×Y ). Assuming, possibly for a subsequence, that γψh → γ, due to the fact that π0# (γψh ) = µ for any h, we obtain that π0# γ = µ; hence, according to Theorem 9.1, we can represent γ as γ = γx ⊗ µ, for suitable probability measures γx in Y , with x → γx Borel. The family of measures γx is called Young limit of the sequence (ψh ); once the Young limit is known, we can compute the w∗ -limit of γ(ψh ) in L∞ (X, µ) for any γ ∈ C(Y ): indeed, using test function of the form φ(x)γ(y), we easily obtain that the limit (in the dual of C(X) and therefore in L∞ (X, µ)) is given by
Lγ (x) := γ(y) dγx (y). Y

We will use the following two well known results of the theory of Young measures. Theorem 9.3 (Approximation theorem). Let γ ∈ M+ (X ×Y ) and write γ = γx ⊗ µ with µ = π0# γ. Then, if µ has no atom, there exists a sequence of Borel maps ψh : X → Y such that γx ⊗ µ = lim δψh (x) ⊗ µ. h→∞

Moreover, we can choose ψh in such a way that the measures ψh# µ have no atom as well. Proof. (Sketch) We assume first that γx = γ is independent of x. By approximation we can also assume that γ=

p

pi δ y i ,

i=1

for suitable yi ∈ Y and pi ∈ [0, 1]. Let Qh , h ∈ Zn , be a partition of Rn in cubes with side length 1/h; since µ has no atom, by Lyapunov theorem we can find a partition Xh1 , . . . , Xhp of X ∩ Qh such that µ(Xhi ) = pi µ(X ∩ Qh ). Then we define ψh = pi on Xhi . For any φ ∈ C(X) and ϕ ∈ C(Y ) we have
X×Y

φϕ d(δψh ⊗ µ) = ∼

p



h∈Zn i=1 p



h∈Zn i=1

and this proves that


ϕ(yi )

φ dµ i Xh




pi ϕ(yi )


φϕ dµ × γ

φ dµ = X∩Qh

X×Y

Lecture Notes on Optimal Transport Problems

49

µ × γ = lim δψh ⊗ µ. h→∞

If we want ψh# µ to be non atomic, the above construction needs to be modified only slightly: it suffices to take small balls Bi centered at yi and to define ψh equal to φi on Xhi , where φi : Rn → Bi is any Borel and one to one map. If γx is piecewise constant (say in a canonical subdivision of X induced by a partition in cubes), then we can repeat the local construction above in each region where γx is constant. Moreover we can approximate Lipschitz functions γx (with respect to the 1-Wasserstein distance) with piecewise constant ones. Finally, any Borel map γx can be approximated by Lipschitz ones through a convolution. We will also need the following result. Lemma 9.1. Let ψh , ψ : X → Y be Borel maps and µ ∈ M1 (X). Then ψh → ψ µ-a.e. if and only if γh := δψh (x) ⊗ µ(x) → γ := δψ(x) ⊗ µ(x) weakly in M(X × Y ). Proof. Assume that ψh → ψ µ-a.e. Since

ϕ dγϕ = ϕ(x, φ(x)) dµ(x) X×Y

∀ϕ ∈ C(X × Y ),

X

the dominated convergence theorem gives that γh → γ. To prove the opposite implication, fix ε > 0 and a compact set K such that ψ|K is continuous and µ(X \ K) < ε. We use as test function ϕ(x, y) = χK (x)γ (y − ψ(x)) , with γ(t) = 1 ∧ |t|/ε (by approximation, although not continuous in X × Y , this is an admissible test function) to obtain lim µ ({x ∈ K : |ψh (x) − ψ(x)| ≥ }) = 0.

h→∞

The conclusion follows letting ε → 0+ . With a similar proof one can obtain a slightly more general result, namely γh := δψh (x) ⊗ µh → γ := δψ(x) ⊗ µ implies ψh → ψ µ-a.e., provided |µh − µ|(X) → 0.

50

Luigi Ambrosio

References [1] [2] [3] [4] [5] [6] [7]

[8] [9] [10] [11] [12]

[13]

[14]

[15]

[16] [17]

[18] [19]

[20]

G.Alberti: On the structure of the singular set of convex functions. Calc. Var. Partial Differential Equations, 2 (1994), 17–27. G.Alberti & L.Ambrosio: A geometric approach to monotone functions in Rn . Math. Z., 230 (1999), 259–316. G.Alberti, B.Kirchheim & D.Preiss: Personal communication. L.Ambrosio, N.Fusco & D.Pallara: Functions of Bounded Variation and Free Discontinuity Problems. Oxford University Press, 2000. L.Ambrosio & B.Kirchheim: Rectifiable sets in metric and Banach spaces. Math. Ann., 318 (2000), 527–555. L.Ambrosio & B.Kirchheim: Currents in metric spaces. Acta Math., 185 (2000), 1–80. J.Benamou & Y.Brenier: A computational fluid mechanics solution to the Monge–Kantorovich mass transfer problem. Numerische Math., 84 (2000), 375–393. Y.Brenier: A homogenized model for vortex sheets. Arch. Rational Mech. Anal., 138 (1977), 319–353. Y. Brenier: D´ecomposition polaire et r´ earrangement monotone des champs de vecteurs. Comm. Pure Appl. Math., 44 (1991), 375–417. Y.Brenier: Polar factorization and monotone rearrangement of vector-valued functions. Arch. Rational Mech. Anal., 122 (1993), 323–351. Y.Brenier: A Monge–Kantorovich approach to the Maxwell equations. In preparation. ´, G.Buttazzo & P.Seppecher: Energies with respect to a G.Bouchitte measure and applications to low dimensional structures. Calc. Var. Partial Differential Equations, 5 (1997), 37–54. ´, G.Buttazzo & P.Seppecher: Shape optimization via MonG.Bouchitte ge–Kantorovich equation. C. R. Acad. Sci. Paris S´er. I Math., 324 (1997), 1185–1191. ´ & G.Buttazzo: Characterization of optimal shapes and G.Bouchitte masses through Monge–Kantorovich equation. J. Eur. Math. Soc. (JEMS), 3 (2001), 139–168. L.Caffarelli, M.Feldman & R.J.McCann: Constructing optimal maps for Monge’s transport problem as a limit of strictly convex costs. J. Amer. Math. Soc., 15 (2002), 1–26. C.Castaing & M.Valadier: Convex analysis and measurable multifunctions. Lecture Notes in Mathematics 580, Springer, 1977. A.Cellina & S.Perrotta: On the validity of the maximum principle and of the Euler–Lagrange equation for a minimum problem depending on the gradient. SIAM J. Control Optim., 36 (1998), 1987–1998. C.Dellacherie & P.Meyer: Probabilities and potential. Mathematical Studies, 29, North Holland, 1978. L.De Pascale & A.Pratelli: Regularity properties for Monge transport density and for solutions of some shape optimizazion problem. Calc. Var. Partial Differential Equations, 14 (2002), 249–274. L.C.Evans: Partial Differential Equations. Graduate Studies in Mathematics, 19, AMS, 1998.

Lecture Notes on Optimal Transport Problems

51

[21] L.C.Evans: Partial Differential Equations and Monge-Kantorovich Mass Transfer. Current developments in mathematics, 65–126, Int. Press. Boston, MA, 1999. [22] L.C.Evans & W.Gangbo: Differential Equation Methods for the MongeKantorovich Mass Transfer Problem. Mem. Amer. Math. Soc., 137 (1999), no. 653. [23] K.J.Falconer: The geometry of fractal sets. Cambridge University Press, 1985. [24] H.Federer: Geometric measure theory. Springer, 1969. [25] M.Feldman & R.J.McCann: Uniqueness and transport density in Monge’s mass transport problem. Forthcoming. [26] W.Gangbo & R.J.McCann: The geometry of optimal transportation. Acta Math., 177 (1996), 113–161. [27] R.Jerrard & H.M.Soner: Functions of bounded higher variation. Indiana Univ. Math. J., 51 (2002), 645–677. [28] D.G.Larman: A compact set of disjoint line segments in R3 whose end set has positive measure. Mathematika, 18 (1971), 112–125. [29] R.Jordan, D.Kinderlehrer & F.Otto: The variational formulation of the Fokker–Planck equation. SIAM J. Math. Anal., 29 (1998), 1–17. [30] F.Morgan: Geometric measure theory – A beginner’s guide. Second edition, Academic Press, 1995. ¨ller: Variational methods for microstructure and phase transitions. [31] S.Mu Proc. CIME summer school “Calculus of variations and geometric evolution problems”, Cetraro (Italy), 1996, S.Hildebrandt and M.Struwe eds, 85–210, Lecture Notes in Math., 1713, Springer, Berlin, 1999. [32] F.Otto: The geometry of dissipative evolution equations: the porous medium equation., Comm. Partial Differential Equations, 26 (2001), 101–174. [33] F.Poupaud & M.Rascle: Measure solutions to the linear multi-dimensional transport equation with nonsmooth coefficients. Comm. Partial Differential Equations, 22 (1997), 301–324. [34] A.Pratelli: Regolarit` a delle misure ottimali nel problema di trasporto di Monge–Kantorovich. Thesis, Pisa University, 2000. ¨schendorf: Mass transportation problems. Probability [35] S.T.Rachev & L.Ru and its applications, Springer, 1998. ¨schendorf: On c-optimal random variables. Statistics & Probability [36] L.Ru Letters, 27 (1996), 267–270. [37] W.Rudin: Real and Complex Analysis. McGraw–Hill, New York, 1974. [38] L.Simon: Lectures on geometric measure theory. Proc. Centre for Math. Anal., Australian Nat. Univ., 3, 1983. [39] S.K.Smirnov: Decomposition of solenoidal vector charges into elementary solenoids and the structure of normal one-dimensional currents. St. Petersburg Math. J., 5 (1994), 841–867. [40] C.Smith & M.Knott: On the optimal transportation of distributions. J. Optim. Theory Appl., 52 (1987), 323–329. [41] C.Smith & M.Knott: On Hoeffding–Fr´echet bounds and cyclic monotone relations. J. Multivariate Anal., 40 (1992), 328–334. [42] V.N.Sudakov: Geometric problems in the theory of infinite dimensional distributions. Proc. Steklov Inst. Math., 141 (1979), 1–178. [43] N.S.Trudinger & X.J.Wang: On the Monge mass transfer problem. Calc. Var. Partial Differential Equations, 13 (2001), 19–31.

52

Luigi Ambrosio

[44] B.White: Rectifiability of flat chains. Ann. of Math. (2), 150 (1999), 165–184. ˇek: On the differentiation of convex functions in finite and infinite [45] L.Zaj´ıc dimensional spaces. Czechoslovak Math. J., 29 (104) (1979), 340–348.

Numerical Approximation of Mean Curvature Flow of Graphs and Level Sets Klaus Deckelnick1 and Gerhard Dziuk2 1

2

Institut f¨ ur Analysis und Numerik, Otto-von-Guericke-Universit¨ at Magdeburg, Universit¨ atsplatz 2, Postfach 4120, D-39106 Magdeburg, Germany [email protected] Institut f¨ ur Angewandte Mathematik, Universit¨ at Freiburg, Hermann-Herder-Straße 10, D-79104 Freiburg, Germany [email protected]

Summary. The mean curvature flow problem for graphs is closely related to the mean curvature of level sets. We design algorithms for both problems in the context of finite elements together with a semi implicit time discretization. The time discretization linearises the problem and is to the effect that in every time step a linear system of equations has to be solved numerically. We prove convergence estimates of optimal order in space and time for the geometric quantities. The results are true in any space dimension. We also provide numerical tests.

1 Introduction The subject of this paper is the numerical solution of the mean curvature flow problem for graphs and for level sets. Mean curvature flow means the motion of a surface or an interface with a velocity, which is its mean curvature. In short, this problem can be written as V = −H. V is the normal velocity and H is the mean curvature of a surface. The normalisation is chosen in such a way that the normal velocity of a sphere is positive and spheres shrink under this law of motion. For graphs the problem can be written as a differential equation for the height function. The resulting differential equation 1



1 + |∇u|2

1 ∂u ∂  ∂u   − =0 ∂t ∂xi 1 + |∇u|2 ∂xi i=1 n

has some unpleasant properties, especially when one is interested in computational methods for its solution. It is highly nonlinear, not uniformly parabolic and it is not in divergence form. Nevertheless there is a quite complete analysis for this problem, some of which will be presented in Section 2.2. The analysis uses the maximum principle and is therefore based on the L∞ -norm.

L. Ambrosio et al.: LNM 1812, P. Colli and J.F. Rodrigues (Eds.), pp. 53–87, 2003. c Springer-Verlag Berlin Heidelberg 2003


54

Klaus Deckelnick and Gerhard Dziuk

We are interested in deriving numerical schemes for this differential equation and in proving convergence of the discrete solutions to the continuous solution under realistic assumptions. As a numerical method we choose finite elements because of their efficiency and accessibility to modern adaptive methods. The finite element method by its nature is a method for reflexive spaces, in particular for spaces which are based on L2 . Although the equation is not in divergence form we shall derive a finite element method for the discretization with respect to space. The time discretization is such that it linearizes the equations. At the end, in every time step a linear system of equations has to be solved. This can be done efficiently by some conjugate gradient algorithm. The numerical schemes will be such that they can be incorporated easily into existing finite element codes for the solution of linear parabolic equations. The error estimates which we shall prove are of a form, which is adapted to the geometric origin of the equation. We shall prove that normal velocity and normal vector converge of optimal order V − Vhτ L2 (0,T ;L2 (Γhτ )) + ν − νhτ L∞ (0,T ;L2 (Γhτ )) ≤ c(τ + h). Here Vhτ is the normal velocity of the discrete surface Γhτ and νhτ is its normal. The index h stands for the spatial grid size and τ stands for the time step size. A quite astonishing result is that we do not have any coupling of τ and h although our time discretization is not implicit, but semi-implicit only. The level set form of mean curvature flow, 1 ∂u ∂  1 ∂u  − = 0, |∇u| ∂t ∂xi |∇u| ∂xi i=1 n

for the scalar level function u = u(x, t) is closely related to the the mean curvature flow problem for graphs via a regularization. By a simple scaling trick the algorithms and convergence results for graphs can be transformed to the level set problem. We prove convergence of the discrete solution of the regularised problem to the viscosity solution of the level set flow. The optimal coupling of discretization parameters τ , h and regularization parameter is an open question. Besides the general convergence result for viscosity solutions we can give numerical examples which suggest that there might be a nontrivial coupling condition. The results for mean curvature flow can be generalized to anisotropic mean curvature flow, see [6]. Although we shall only treat the case of piecewise linear finite elements a generalization to higher order C 0 -elements is straightforward. It creates technical difficulties at the boundary and higher regularity assumptions on the continuous solution. Fried [14] used the level set finite element method for the computation of dendritic growth. Further information about numerical level set methods are contained in [18], [19] and [21].

Numerical Approximation of Mean Curvature Flow

55

Notations H m,p (Ω) denotes the usual Sobolev space. The corresponding norm is given by uH m,p (Ω) =

m 

Dk upLp (Ω)

 p1

,

k=0

with the usual modification for p = ∞. For p = 2 we simply write H m (Ω) = H m,2 (Ω) with norm ·H m (Ω) ; furthermore we use · to denote the L2 -norm. We shall often use, if not otherwise indicated, v m (x) = v(x, mτ ) as an abbreviation for the evaluation of a function v = v(x, t) on the m-th time level t = mτ for given time step size τ > 0.

2 Mean Curvature Flow of Graphs 2.1 The Differential Equation We study the numerical computation of the isotropic evolution of n-dimen– sional surfaces which are graphs in Rn+1 . The law of motion is given by V = −H.

(1)

Here V represents the normal velocity and H is the mean curvature of the graph Γ (t) = {(x, u(x, t)) | x ∈ Ω} over a bounded domain Ω ⊂ Rn . The normal to Γ is chosen to be ν(u) =

 (∇u, −1) , Q(u) = 1 + |∇u|2 . Q(u)

The normal velocity of Γ then can be written as V (u) = −

ut . Q(u)

(2)

The mean curvature of Γ is given by H(u) = ∇ · 

∇u 1 + |∇u|2

.

Equation (1) then leads to the differential equation ut



1+

|∇u|2

−∇· 

∇u 1 + |∇u|2

= 0.

(3)

56

Klaus Deckelnick and Gerhard Dziuk

We shall solve the initial boundary value problem ut −



1 + |∇u|2 ∇ · 

∇u

= 0 in Ω × (0, T ) 1 + |∇u|2 u = u0 on ∂Ω × (0, T ) ∪ Ω × {0}.

(4)

The boundary data are chosen independently of time. A generalization to time dependent boundary values is possible and adds some technical problems only. In more explicit form the differential equation reads ut −

n 

δij −

i,j=1

uxi uxj 1 + |∇u|2

 uxi xj = 0.

(5)

The equation is parabolic but uniformly parabolic only if |∇u| is uniformly bounded. With uxi uxj aij = δij − 1 + |∇u|2 one has, for arbitrary ξ ∈ Rn , n

i,j=1

aij ξi ξj = |ξ|2 −

  2 1 |∇u|2 (∇u · ξ) 2 1 − ≥ |ξ|2 ≥ |ξ| . 1 + |∇u|2 1 + |∇u|2 1 + |∇u|2

We also observe that the equation is not in divergence form. 2.2 Some analysis for the problem Analytical results for the problem 4 are due to Huisken [15] and Lieberman [16]. Theorem 2.1. Let u0 be from C 2,α (Ω) and let the domain Ω be mean convex with ∂Ω ∈ C 2,α . Assume that the function u0 satisfies the compatibility condition H(u0 ) = 0 on ∂Ω. Then there exists a unique solution α u ∈ C 2+α,1+ 2 (Ω × [0, T ]) of problem (4) for all T < ∞. The assumption that the boundary of the domain has nonnegative mean curvature is a necessary condition. If it is not satisfied the continuous solution breaks down after finite time, although, as we shall see later, the discrete solution exists for all times. The main tools in the proof of the previous theorem are a boundary gradient estimate and an evolution equation for the area element Q = Q(u). The important fact in (6) below is, that the terms on the right hand side of the differential equation are negative and the maximum principle is available for an estimate of Q.  Lemma 2.1. Let u be a smooth solution of (4). Then with Q = 1 + |∇u|2 and νi = uxi /Q (i = 1, . . . , n) the differential equation

Numerical Approximation of Mean Curvature Flow

Qt − =−

n

57

(δij − νi νj )Qxi xj

i,j=1 n

n 1 1

(δij − νi νj )Qxi Qxj − (δij − νi νj )uxm xi uxm xj Q i,j=1 Q i,j,m=1

(6)

holds in Ω × (0, T ). Later our numerical algorithms will be based on a variational formulation of the problem, i.e. on an L2 (Ω) formulation. For that purpose integral estimates for the solution will be important. Also we will not have the maximum principle as a tool for our discrete equations and therefore will have to estimate L2 (Ω)-norms. Lemma 2.2. For the solution of problem (4) one has the energy equation

u2t d + Q(u) = 0. (7) dt Ω Ω Q(u) If u0 = 0 on ∂Ω, then


d |∇u|2 u2t u2 2 + = 0. u Q(u) + 2 dt Ω Ω Q(u) Ω Q(u)

(8)

For later use it is necessary to identify the important geometric quantities in the energy equations. The equations (7) and (8) can be written as
d V (u)2 + |Γ | = 0, dt Γ


d 2 2 u +2 |∇u| + V (u)2 u2 = 0, dt Γ Γ Γ where ∇ denotes the tangential gradient on Γ . Proof (of Lemma 2.2). For the first equation (7) we observe that, since ut = 0 on ∂Ω,



u2t d ∇u ∇u = =− =− ut ∇ · ∇ut · Q(u). Q(u) Q(u) dt Ω Ω Q(u) Ω Ω The second equation uses the differential equation twice.

58

Klaus Deckelnick and Gerhard Dziuk

1 d 2 dt





∇u 1 · ∇ut u2 2 Ω Q(u) Ω


∇u |∇u|2 1 · ∇(u2 ut ) − = uut Q(u) + uut 2 Q(u) Q(u) Ω
Ω
Ω uut ∇u 1 = + · ∇(u2 ut ) 2 Ω Q(u) Ω Q(u)

1 ∇u ∇u 2 = − u ut u∇ · ∇· Q(u) 2 Q(u) Ω Ω

|∇u|2 u2 u2t 1 − . =− 2 Ω Q(u) Ω Q(u)


u2 Q(u) = Ω

uut Q(u) +

This is the second energy equation (8). Our error analysis for the numerical approximation of (4) will be carried out under the assumption that (4) has a solution u which satisfies N := sup uH 2,∞ (Ω) + sup ut H 1,∞ (Ω) (0,T )




(0,T )




T

ut 2H 2 (Ω) +

+ 0

T

utt 2 < ∞.

(9)

0

An existence theorem in Sobolev spaces is provided in [3]. 2.3 Discretization Discretization in Space Let Th be a family of triangulations of Ω with maximum mesh size h := maxS∈Th diam(S). We denote by Ωh the corresponding discrete domain, i.e.  ¯h = S Ω S∈Th

and assume that all vertices on ∂Ωh also lie on ∂Ω. Furthermore we suppose that the triangulation is nondegenerate in the sense that max

S∈Th

diam(S) ≤ κ, ρS

where the constant κ > 0 is independent of h and ρS denotes the radius of ¯ The discrete space is the largest ball which is contained in the simplex S. defined by ¯h ) | vh is a linear polynomial on each S ∈ Th } Xh := {vh ∈ C 0 (Ω and Xh0 := Xh ∩ H01 (Ωh ).

Numerical Approximation of Mean Curvature Flow

59

There exists an interpolation operator Ih : H 2 (Ωh ) → Xh , mapping H (Ωh ) ∩ H01 (Ω) into Xh0 , such that 2

v − Ih v + h∇(v − Ih v) ≤ ch2 vH 2 (Ωh )

(10)

for all v ∈ H (Ωh ). Although the differential equation is not in divergence form, one can easily derive a variational form of (4), namely

ut φ ∇u · ∇φ + = 0, φ ∈ H01 (Ω), 0 < t < T. (11) Q(u) Ω Q(u) Ω 2

The semi–discrete approximation of (11) then is given by the following Algorithm. Algorithm 1 Let u0h = Ih u0 . For t ∈ [0, T ) find uh (·, t) ∈ Xh , with uh (·, t)− Ih u0 ∈ Xh0 and uh (·, 0) = uh0 = Ih u0 , such that

uht φh ∇uh · ∇φh + =0 (12) Q(u ) Q(uh ) h Ωh Ωh holds for all t ∈ (0, T ) and all discrete test functions φh ∈ Xh0 . Fully discrete schemes and stability The derivation of fully discrete schemes for the mean curvature flow problem is an important task for applications. We have to prove stability and convergence and have to show the efficiency of the algorithm. In the following we list three different discretizations of (12) with respect to time. We use the notation v m (x) = v(x, mτ ) for m = 0, . . . , M , with time step τ > 0 and M ≤ [T /τ ]. A fully implicit scheme for (4) is given by

(um+1 − um ∇um+1 · ∇φh 1 h )φh h h + =0 m+1 m+1 τ Ωh Q(uh ) Q(uh ) Ωh (m = 0, . . . , M − 1) for all φh ∈ Xh0 . So, in every time step one has to solve a highly nonlinear elliptic problem. The numerically simplest time discretization is the explicit one,

(um+1 − um 1 ∇um h )φh h · ∇φh h + =0 m) m τ Ωh Q(uh ) Q(u Ωh h (m = 0, . . . , M − 1), for all φh ∈ Xh0 , for which numerically one only has to invert the mass matrix (weighted by Q(um h )). If one uses mass lumping, every time step only requires one matrix multiplication. However, just as for the linear heat equation, a severe restriction on the smallness of the time step size with respect to the grid size h will be necessary. It then certainly is worth to try some semi–implicit scheme for time discretization. We shall discuss the scheme in full detail in Section 2.4.

60

Klaus Deckelnick and Gerhard Dziuk

Algorithm 2 Let u0h = Ih u0 . For m = 0, . . . M − 1 compute um+1 ∈ Xh h such that uh − u0h ∈ Xh0 and for every φh ∈ Xh0 1 τ


Ωh

um+1 φh h + m Q(uh )




∇um+1 · ∇φh 1 h = m τ Q(uh )

Ωh


Ωh

um h φh . Q(um h )

(13)

Here in every time step one has to solve a discrete linear elliptic differential equation with stiffness matrix weighted by Q(um h ). As a stability criterion we use the basic energy norm introduced in (7). The fact that we have unconditional stability, although the scheme is not fully implicit but semi-implicit in time, is a first hint that we might expect unconditional convergence. See Section 2.4 for the details. Theorem 2.2. The solution um h , 0 ≤ m ≤ M of (2) satisfies, for every m ∈ {1, . . . , M }, τ

m−1


k=0

+

|Vhk |2 Q(ukh ) +

Ωh

m−1


k=0

Ωh

k 2 (Q(uk+1 h ) − Q(uh ))

1 Q(ukh )



m−1
1

k+1 k 2 m |ν(uk+1 ) − ν(u )| Q(u ) + Q(u ) = Q(u0h ), (14) h h h h 2 Ωh Ωh Ωh k=0

where Vhk = −

uk+1 − ukh h τ Q(ukh )

is the discrete normal velocity. Proof. We choose φh = uk+1 − ukh as a test function in (2) for m = k and get h 1 τ


Ωh

(uk+1 − ukh )2 h + Q(ukh )

Let us use the notation ν(vh ) =


Ωh

∇uk+1 · ∇(uk+1 − ukh ) h h = 0. Q(ukh )

∇vh Q(vh ) ,

(15)

−1 so that ν(vh ) = (ν(vh ), Q(v ). Then h)

2 Q(uk+1 · ∇(uk+1 − ukh ) ∇uk+1 k+1 k h h h ) −1 = − ν(uk+1 h ) · ν(uh )Q(uh ) k k Q(uh ) Q(uh )

=

2 Q(uk+1 1 k+1 k+1 k 2 h ) + |ν(uk+1 h ) − ν(uh )| Q(uh ) − Q(uh ) k 2 Q(uh )

k 2 (Q(uk+1 1 k+1 k+1 k 2 k h )−Q(uh )) . = |ν(uk+1 h )−ν(uh )| Q(uh )+Q(uh )−Q(uh )+ 2 Q(ukh )

We use this result in (15) and sum over k = 0, . . . , m − 1.

Numerical Approximation of Mean Curvature Flow

61

The continuous solution of (4) exists globally in time only if the domain Ω is mean convex. In contrast to this, the discrete solution um h exists for m ∈ N, i.e. for T = ∞ without any geometric condition on the domain Ωh . Lemma 2.3. The semi-discrete solution uh from Algorithm 1 and the discrete solution um h from Algorithm 2 exist for all times. Proof. This is a simple consequence of the fact that area is bounded uniformly. From Theorem 2.2 we have that

sup Q(um ) ≤ Q(u0h ) h m∈N

Ωh

Ωh

and this implies that −n sup sup Q(um h ) ≤ ch

m∈N Ωh


Q(u0h ). Ωh

Thus the linear system (13) is uniquely solvable for every m ∈ N. A similar argument holds for the semi-discrete algorithm. In Figure 1 we show the discrete solution of (4) on a non convex smooth domain in the plane. Local existence of a smooth solution is known. Under certain parameters the gradient blows up near the non convex part of the boundary and forms a “free boundary” there [17]. Since the discrete solution exists for all times, the gradient at the boundary in the numerical results only becomes “nearly vertical”. Convergence of the semi-discrete scheme In [4] we proved convergence for the semi-discrete Finite Element scheme from Algorithm 1, see also [3]. Here we only give the result. The estimates depend on the quantity  
T 2 2 sup u(., t)H 2,∞ (Ω) + ut (., t)H 1,∞ (Ω) + ∇ut 2H 2 (Ω) , (16) (0,T )

0

which is finite for the continuous solution under suitable assumptions on the data, see [4]. We assume that we have a solution of this quality up to time T . The error estimate in the next theorem then is valid as long as this solution exists. Theorem 2.3. Let u be a solution of the continuous problem (4) with finite norms (16). Then there exists a unique solution uh ∈ C 1 ((0, T ), Xh ) of the semi-discrete problem (12) and


T





2

(V (u) − V (uh ))2 Q(uh ) + sup 0

Ω∩Ωh

(0,T )

|ν(u) − ν(uh )| Q(uh ) ≤ ch2 . Ω∩Ωh

The constant c depends on the norms (16).

62

Klaus Deckelnick and Gerhard Dziuk

Fig. 1. Evolution of a graph over a non convex domain by mean curvature flow

Numerical Approximation of Mean Curvature Flow

63

2.4 Convergence of the fully discrete semi implicit scheme For domains in the plane and under the condition τ ≤ δ0 h, for some constant δ0 > 0, the convergence of this algorithms is proved and optimal error estimates are given in [5]. In this section we are going to give a proof of the unconditional convergence together with optimal error estimates for the geometric quantities V and ν of the fully discrete semi-implicit scheme (12). Thus no coupling of time step size and spatial grid size is necessary although the algorithm is not implicit. We emphasize that the results are true for any space dimension. ¯ ∪ Ωh for all h ≤ 1. In view Choose on open set Ω  ⊂ Rn which contains Ω of the regularity (9) of u and since ∂Ω is smooth, there exists an extension u ¯ : Ω  × [0, T ] → R such that u ¯|Ω×[0,T ] = u and
T
T uH 2,∞ (Ω
) + sup ¯ ut H 1,∞ (Ω
) + ¯ ut 2H 2 (Ω
) + ¯ utt 2 ≤ cN, sup ¯ (0,T )

(0,T )

0

0

(17) where N appeared in (9). Theorem 2.4. Assume that there exists a solution of (4) on (0, T ) × Ω and that the norms (9) are finite and let um h , (m = 1, . . . , M ) be the solution of Algorithm 2. Then there exists a τ0 > 0 such that for all 0 < τ ≤ τ0 τ

M


m=0




sup m=0,...,M

2 2 (V (um ) − Vhm )2 Q(um h ) ≤ c(τ + h ),

(18)

2 m 2 2 |ν(um ) − ν(um h )| Q(uh ) ≤ c(τ + h ),

(19)

Ω∩Ωh

Ω∩Ωh

where M = [ Tτ ] and Vhm = −

um+1 − um 1 h h τ Q(um h )

is the discrete normal velocity. The proof of the Theorem will be based on some Lemmas, which we are going to prove now. In the first Lemma we provide the error relation by subtracting suitable forms of the discrete and the continuous equation. Lemma 2.4. For every φh ∈ Xh0 we have the error relation  
 m+1
 um+1 ∇um+1 − um ∇um+1 1 u − um h h h − φh + − · ∇φh τ Ωh Q(um ) Q(um Q(um ) Q(um Ωh h ) h )  

1 ∇(um+1 − um ) um+1 − um · ∇φ − um φh + = h t m Q(u ) τ Q(um ) Ωh Ωh
− (H(um ) + V (um )) φh . (20) Ωh \Ω

64

Klaus Deckelnick and Gerhard Dziuk

Proof. For every φ ∈ H01 (Ωh ) we have that

∇um ∇um · ∇φ = − φ ∇ · m Q(um ) Ωh Q(u ) Ω

h ∇um ∇um φ − φ ∇· ∇ · =− Q(um ) Q(um ) Ω∩Ωh Ωh \Ω

um ∇um t =− φ − φ ∇ · m Q(um ) Ω∩Ωh Q(u ) Ωh \Ω

um+1 − um 1 um+1 − um 1 =− φ + − um t )φ m m ( τ Ωh Q(u ) τ Ωh Q(u )
um ∇um t )φ (∇ · − m − Q(u ) Q(um ) Ωh \Ω and this leads to the equation

um+1 − um ∇um 1 φ + m m · ∇φ τ Ωh Q(u ) Ωh Q(u )

1 um+1 − um um ∇um t m = − ut )φ − )φ. (21) (∇ · m ( m − τ Q(u ) Q(um ) Ωh Q(u ) Ωh \Ω The difference of equations (13) and (21) gives the error relation (20). The following Lemma contains the key estimate for the proof of Theorem 2.4. With this result it will be possible to estimate the L2 -norm (on the discrete surface) of the error with respect to the normals, ν(u(·, τ m))−ν(um h ), uniformly in time. For this we provide an estimate for the discrete time derivative of this quantity. Here and in the following we use the abbreviations Qk = Q(uk ), Qkh = Q(ukh ), ν k = ν(uk ), νhk = ν(ukh ),

(22)

for k = m, m + 1. Lemma 2.5. 
 ∇um ∇um h − · ∇((um+1 − um+1 ) − (um − um h )) h Q(um ) Q(um ) Ωh h

1 1 m+1 2 2 m = |ν(um+1 ) − ν(um+1 )| Q(u ) − |ν(um ) − ν(um h )| Q(uh ) h h 2 Ωh 2 Ωh

1 m+1 m m 2 − |(ν(um+1 ) − ν(um+1 )) − (ν(u ) − ν(u ))| Q(u ) + Rm , h h h 2 Ωh Ωh (23)

Numerical Approximation of Mean Curvature Flow

65

with m+1 ) − Q(um+1 )) Rm = (ν(um+1 ) − ν(um )) · (ν(um ) − ν(um h ))(Q(u h 1 2 m+1 ) − Q(um )). + |ν(um ) − ν(um h )| (Q(u 2

Proof. We use the relations (∇v, −1) = ν(v)Q(v),

ν(v) · (ν(v) − ν(w)) =

1 |ν(v) − ν(w)|2 2

and obtain   ∇um ∇um h − m · ∇((um+1 − um+1 ) − (um − um h )) h Qm Qh ) − (ν m Qm − νhm Qm = (ν m − νhm ) · ((ν m+1 Qm+1 − νhm+1 Qm+1 h )) h m+1 m+1 m+1 m+1 m+1 m+1 m m m m = (ν −νh )·(ν Q −νh Qh )−(ν −νh )·(ν Q −νhm Qm h ) −((ν m+1 − νhm+1 ) − (ν m − νhm )) · (ν m+1 Qm+1 − νhm+1 Qm+1 ) h 1 1 m+1 − νhm+1 |2 (Qm+1 + Qm+1 ) − |ν m − νhm |2 (Qm + Qm = |ν h ) h 2 2 ) −((ν m+1 − νhm+1 ) − (ν m − νhm )) · (ν m+1 Qm+1 − νhm+1 Qm+1 h 1 1 m+1 ˜m − νhm+1 |2 Qm+1 − |ν m − νhm |2 Qm = |ν h +R , h 2 2

(24)

where ˜ m = 1 |ν m+1 − ν m+1 |2 Qm+1 − 1 |ν m − ν m |2 Qm R h h 2 2 m+1 m+1 m m −((ν − νh ) − (ν − νh )) · (ν m+1 Qm+1 − νhm+1 Qm+1 ). h We rewrite this term as follows ˜ m = − 1 |ν m+1 − ν m+1 |2 Qm+1 − 1 |ν m −νhm |2 Qm R h h 2 2 m m+1 m +(ν −νh ) · (ν − νhm+1 )Qm+1 + (ν m −νhm )·ν m+1(Qm+1 −Qm+1 ) h h 1 1 m+1 m+1 −νh )−(ν m −νhm )|2 Qm+1 + |ν m −νhm |2 (Qm+1 −Qm ) = − |(ν h 2 2 1 m m 2 m+1 m+1 m m ) + (ν − νh ) · ν m+1 (Qm+1 − Qm+1 ) + |ν −νh | (Qh −Q h 2 1 1 + |ν m −νhm |2 (Qm+1 −Qm ) = − |(ν m+1 −νhm+1 )−(ν m −νhm )|2 Qm+1 h 2 2 1 m m m+1 m m+1 m+1 m (Q −Qh ) + (ν −νh )(Qm+1 −Qm+1 )) +(ν −νh ) · (ν h 2 1 1 + |ν m −νhm |2 (Qm+1 −Qm ) = − |(ν m+1 −νhm+1 )−(ν m −νhm )|2 Qm+1 h 2 2 ) +(ν m −νhm ) · (ν m+1 −ν m )(Qm+1 −Qm+1 h 1 m m m+1 m m m m+1 −Qh ) + (ν −νh )(Qm+1 −Qm+1 )). +(ν −νh ) · (ν (Q h 2

66

Klaus Deckelnick and Gerhard Dziuk

The last term vanishes because 1 1 ν m(Qm+1 −Qm+1 )+ (ν m −νhm )(Qm+1 −Qm+1 )=(Qm+1 −Qm+1 ) (ν m +νhm ) h h h 2 2 and we finally have ˜ m = −1 |(ν m+1 −ν m+1 )−(ν m −νhm )|2 Qm+1 + 1 |ν m −νhm |2 (Qm+1 −Qm ) R h h 2 2 m m m+1 m m+1 −ν )(Q −Qm+1 ). +(ν −νh ) · (ν h Together with (24) this proves the Lemma. Lemma 2.6. 
 −um ∇(um+1 −um ) ∇(um+1 h ) h − · ∇((um+1 −um+1 ) − (um −um h )) h m m) Q(u ) Q(u Ωh h
m+1 2 = |(ν(um+1 ) − ν(um+1 )) − (ν(um ) − ν(um ) h ))| Q(uh h Ωh




+ Ωh

(Q(um+1 ) − Q(um ))2 + Q(um )


Ωh

2 (Q(um+1 ) − Q(um h )) h + Q(um h )


S m , (25) Ωh

with S m = |ν(um+1 ) − ν(um )|2 (Q(um+1 ) − Q(um+1 )) h m+1 m+1 m m +2(ν(u ) − ν(u )) · (ν(uh ) − ν(uh ))Q(um+1 ) h Q(um+1 ) Q(um+1 ) h − ν(um )) · (ν(um+1 − ν(um ) h )) h m Q(u ) Q(um ) h (26) ×(Q(um ) + Q(um h )).

−(ν(um+1 )

Proof. We use the relation |∇(v − w)|2 = |ν(v) − ν(w)|2 Q(v)Q(w) + (Q(v) − Q(w))2

(27)

and get   − um ∇(um+1 − um ) ∇(um+1 h ) h · ∇((um+1 − um+1 + ) − (um − um h )) h Qm Qm h =

2 |∇(um+1 − um |∇(um+1 − um )|2 h )| h − m m Q Qh   1 1 m −∇(um+1 − um ) · ∇(um+1 − u ) + h h Qm Qm h

(Qm+1 −Qm )2 +|νhm+1 −νhm |2 Qm+1 h Qm (Qm+1 −Qm )2 Qm+1 Qm+1 + h m h −(ν m+1 m −ν m ) · (νhm+1 h m −νhm )(Qm +Qm h ) Qh Q Qh

= |ν m+1 −ν m |2 Qm+1 +

Numerical Approximation of Mean Curvature Flow

= |(ν m+1 −ν m )−(νhm+1 −νhm )|2 Qm+1 + h

67

−Qm )2 (Qm+1 −Qm )2 (Qm+1 + h m h m Q Qh

+|ν m+1 − ν m |2 (Qm+1 − Qm+1 ) + S˜m , h with S˜m = 2(ν m+1 − ν m ) · (νhm+1 − νhm )Qm+1 h −(ν m+1

m+1 Qm+1 m+1 Qh m − ν ) · (ν − νhm )(Qm + Qm h ). h Qm Qm h

We are now prepared to prove the convergence Theorem for the fully discrete scheme from Algorithm 2. Proof of Theorem 2.4. We insert φh = Ih (um+1 −um )−(um+1 −um h ) h m+1 m+1 m m+1 = ((u −u )−(uh −um −um )−Ih (um+1 −um )) h ))−((u = (em+1 −em )−(εm+1 −εm ), as a test function into (20), where we set e = u − uh , ε = u − Ih u, and rephrase the result as A1 + A2 + A3 = A4 + A5 + A6 + A7 + A8 , with

 um+1 − um um+1 − um h h − ((em+1 − em ) − (εm+1 − εm )) m m) Q(u ) Q(u Ωh h   ∇um+1 ∇um+1 h · ∇((um+1 − um+1 − ) − (um − um h )) h Qm Qm Ωh h 
 ∇um+1 ∇um+1 h · ∇(εm+1 − εm ) − − m m Q Q Ωh h
∇(um+1 − um ) · ∇(em+1 − em ) Qm Ωh
∇(um+1 − um ) − · ∇(εm+1 − εm ) Qm Ωh 
 m+1 1 u − um − um − (εm+1 − εm ) t m τ Q Ωh 
 m+1 1 m+1 u − um m − ut (e − em ) m τ Q Ωh
− (H(um ) + V (um ))((em+1 − em ) − (εm+1 − εm )).

1 A1 = τ
A2 = A3 = A4 = A5 = A6 = A7 = A8 =

(28)






Ωh \Ω

68

Klaus Deckelnick and Gerhard Dziuk

Let us start by estimating the first term A1 on the left hand side of (28). We use the abbreviation Vm =−

um+1 − um um+1 − um m h h , V = − h τ Qm τ Qm h

and have, with Young’s inequality and (10),

A1 = τ (V m − Vhm )2 Qm + τ V m (V m − Vhm )(Qm − Qm h h ) Ωh Ωh
+ (V m − Vhm )(εm+1 − εm ) Ωh

1 ≥ τ 2







(V Ωh

m

−

Vhm )2 Qm h

− cτ V

m

Qm 2L∞ (Ωh )

Ωh

1 −c εm+1 − εm 2L2 (Ωh ) τ

1 m m 2 m ≥ τ (V − Vh ) Qh − cτ |ν m − νhm |2 Qm h 2 Ωh Ωh
(m+1)τ −ch4 ut 2H 2 (Ωh ) ,



1 1 − m Qm Qh

2 Qm h

(29)

mτ

with c depending on ut L∞ (Ωh ×(0,T )) . The most difficult term to estimate is the second one on the right hand side of (28). We use the results of the previous two Lemmas and obtain the identities 
 ∇um ∇um h A2 = · ∇((um+1 −um+1 − )−(um −um h )) h Qm Qm Ωh h 
 −um ∇(um+1 −um ) ∇(um+1 h ) h · ∇((um+1 −um+1 − )−(um −um + h )) h Qm Qm Ωh h

1 1 2 m = |ν(um+1 )−ν(um+1 )|2 Q(um+1 )− |ν(um )−ν(um h )| Q(uh ) h h 2 Ωh 2 Ωh
1 + |(ν m+1 −νhm+1 )−(ν m −νhm )|2 Qm+1 h 2 Ωh



2 (Qm+1 −Qm (Qm+1 −Qm )2 m h ) h + + + R + Sm Qm Qm Ωh Ωh Ωh Ωh h

1 1 m+1 2 m+1 m+1 2 m = |ν(u )−ν(uh )| Q(uh )− |ν(um )−ν(um h )| Q(uh ) 2 Ωh 2 Ωh
1 + |(ν m+1 −νhm+1 )−(ν m −νhm )|2 Qm+1 h 2 Ωh

2 (Qm+1 −Qm+1 )−(Qm −Qm h )) h + + T m, (30) Qm Ωh Ω h h

Numerical Approximation of Mean Curvature Flow

69

with the remainder term 1 T m = (ν m+1 −ν m ) · (ν m −νhm )(Qm+1 −Qm+1 )+ |ν m −νhm |2 (Qm+1 −Qm ) h 2 1 1 m+1 m+1 m 2 m+1 m+1 − ν | (Q − Qh ) + (Q − Qm )2 ( m − m ) +|ν Q Qh +2(ν m+1 − ν m ) · (νhm+1 − νhm )Qm+1 h −(ν m+1 +2

Qm+1 Qm+1 − ν m ) · (νhm+1 h m − νhm )(Qm + Qm h ) m Q Qh

(Qm+1 − Qm )(Qm+1 − Qm h ) h . Qm h

Elementary calculations lead to the following form of T m with the special and important feature, that it contains triple products of differences only. 1 )+ |ν m −νhm |2 (Qm+1 −Qm ) T m = (ν m+1 −ν m ) · (ν m −νhm )(Qm+1 −Qm+1 h 2 1 1 ) + (Qm+1 − Qm )2 ( m − m ) +|ν m+1 − ν m |2 (Qm+1 − Qm+1 h Q Qh − Qm+1 ) +(νhm+1 − νhm ) · (ν m+1 − ν m )(Qm+1 h +(ν m+1 − νhm+1 ) · (ν m+1 Qm+1 − ν m Qm )

Qm+1 − Qm h h m Qh

Qm+1 − Qm Qm+1 − Qm 1 1 − |ν m+1 − ν m |2 Qm h m h − |νhm+1 − νhm |2 Qm h 2 Qh 2 Qm +(νhm+1 − ν m+1 ) · (νhm+1 Qm+1 − νhm Qm h ) h +(Qm+1 − Qm )(Qm+1 − Qm h )( h

Qm+1 − Qm Qm

1 1 − m) Qm Q h

m+1 ) +(ν m+1 − ν m ) · (νhm − ν m )(Qm h − Qh m+1 m m m m m+1 +(νh − νh ) · (ν − νh )(Q − Q ) 1 1 m+1 m+1 − ν m |2 (Qm ) − |νhm+1 − νhm |2 (Qm − Qm+1 ). − |ν h − Qh 2 2 In the following we provide a first estimate for the remainder term T m . The constants depend on the norms ∇ut L∞ (Ω×(0,T )) and ∇uL∞ (Ω×(0,T )) .

c|T m | ≤ τ |ν m − νhm ||Qm+1 − Qm+1 | + τ |ν m − νhm |2 + τ 2 |Qm+1 − Qm+1 | h h m+1 2 m m +τ |ν − νh | + τ |νh − νhm ||Qm+1 − Qm+1 | h +τ |ν m+1 − νhm+1 ||νhm+1 − νhm |Qm+1 + τ 2 |νhm+1 − νhm |Qm+1 h h +τ |ν m+1 − νhm+1 |(|νhm+1 − νhm |Qm+1 + |Qm+1 − Qm h |) h h m+1 m m +τ |νhm+1 − νhm |2 Qm − Qm h + τ |ν − νh ||Qh h | m+1 m+1 m m m 2 +τ |ν − νh ||νh − νh | + τ |Qh − Qm | + τ |νhm+1 − νhm |2 . h

70

Klaus Deckelnick and Gerhard Dziuk

We use the simple estimates m m m m | ≤ |(Qm+1 − Qm+1 ) − (Qm − Qm |Qm+1 − Qm+1 h )| + |ν − νh |Qh Q , h h |ν m+1 − νhm+1 | ≤ |(ν m+1 − νhm+1 ) − (ν m − νhm )| + |ν m − νhm |, m+1 |Qm+1 − Qm − Qm+1 ) − (Qm − Qm h | ≤ |(Q h )| + cτ, h h

|νhm+1 − νhm | ≤ |(ν m+1 − νhm+1 ) − (ν m − νhm )| + cτ, Qm h ≤c

2 |(Qm+1 − Qm+1 ) − (Qm − Qm h )| h + Qm+1 + cτ, h Qm h

τ 2 |ν m − νhm | ≤ cτ |ν m − νhm |2 + cτ 3 ,

(31)

together with Young’s inequality and get for every positive number δ, by standard techniques, the estimate |T m | ≤ (δ + cδ τ )

2 |(Qm+1 − Qm+1 ) − (Qm − Qm h )| h m Qh

+(δ + cδ τ )|(ν m+1 − νhm+1 ) − (ν m − νhm )|2 Qm+1 h m+1 3 m +cδ τ |ν m − νhm |2 Qm ). h + cδ τ (Qh + Qh

(32)

Thus from (30) and (32) we get, after a suitable choice of δ and τ0 for τ ≤ τ0 , the estimate 
 ∇um+1 ∇um+1 h A2 = · ∇((um+1 − um+1 − ) − (um − um h )) h m m Q Q Ωh h

1 1 ≥ |ν m+1 − νhm+1 |2 Qm+1 − |ν m − νhm |2 Qm h h 2 Ωh 2 Ωh
1 + |(ν m+1 − νhm+1 ) − (ν m − νhm )|2 Qm+1 h 4 Ωh
2 ((Qm+1 − Qm+1 ) − (Qm − Qm 1 h )) h + 4 Ωh Qm h

m+1 m m 2 m 3 −cτ |ν − νh | Qh − cτ (Qm ). (33) h + Qh Ωh

Ωh

From (31) we infer   ∇um+1 ∇um+1 h | − | Qm Qm h = |ν m+1

m+1 Qm+1 1 1 m+1 Qh − ν + ( m − m )en+1 | h m m Q Qh Q Qh

≤ c|ν m+1 − νhm+1 |

Qm+1 Qm+1 Qm+1 + | m − h m | + |ν m − νhm | m Q Q Qh

Numerical Approximation of Mean Curvature Flow

≤

71

|(Qm+1 − Qm+1 ) − (Qm − Qm h )| h + c|ν m − νhm | Qm h

+c|(ν m+1 − νhm+1 ) − (ν m − νhm )| |(Qm+1 − Qm+1 ) − (Qm − Qm h )| h  m Qh  m+1 − νhm+1 ) − (ν m − νhm )| Qm+1 . +c|ν m − νhm | Qm h h + c|(ν

≤

This then leads to an estimate for A3 
 ∇um+1 ∇um+1 h · ∇(εm+1 − εm ) A3 = − − Qm Qm Ωh h

2 ((Qm+1 − Qm+1 ) − (Qm − Qm h )) h − c τ |ν m − νhm |2 Qm ≥ −δ δ h Qm Ωh Ω h h
(m+1)τ
−δ |(ν m+1 −νhm+1 )−(ν m −νhm )|2 Qm+1 −cδ τ h2 ut 2H 2 (Ωh ) . h Ωh

mτ

(34) Clearly for A5 ,


∇(um+1 − um ) · ∇(εm+1 − εm ) m Q Ωh


12 (m+1)τ 3 2 ≤ cτ 2 h∇ut L∞ (Ωh ×(0,T )) ut H 2 (Ωh )

A5 = −




mτ

(m+1)τ

≤ ch2 mτ

ut 2H 2 (Ωh ) + cτ 3 .

(35)

Let us now estimate the first term A4 on the right hand side.

∇(um+1 −um ) ∇(um+1 −um ) m+1 m m+1 m · ∇(e −e ) = − ∇· (e −e ) m Q Qm Ωh Ωh
1 m+1 =− − um )(em+1 − em ) m ∆(u Ωh Qh 
 1 1 − ∆(um+1 − um )(em+1 − em ) − m m Q Q Ω h
h 1 − ∇ m · ∇(um+1 − um )(em+1 − em ) Q Ωh

72

Klaus Deckelnick and Gerhard Dziuk


≤

|∆((um+1 − um )| Ωh




|em+1 − em | Qm h

|ν m − νhm ||∆((um+1 − um )||em+1 − em |

+ Ωh




|∇(um+1 − um )||em+1 − em |

+c Ωh

=: I1 + I2 + I3 . Let us estimate these three terms separately. For all of them we use the inequality m m m |em+1 − em | ≤ cτ |V m − Vhm |Qm h + cτ |ν − νh |Qh .

(36)

Thus, we have
(m+1)τ  I1 ≤ cτ ut H 2 (Ωh ) V m − Vhm L2 (Ωh ) + ν m − νhm L2 (Ωh ) mτ


≤ δτ







(V m −Vhm )2 Qm h +cτ Ωh

(m+1)τ

2 |ν m −νhm |2 Qm h +cδ τ Ωh

ut 2H 2 (Ωh ) .

mτ

With (36) we get 


Ωh

νhm |2 Qm h

 12 


(em+1 − em )2 Qm h

|ν − I2 ≤ cuL∞ ((0,T ),H 2 (Ωh )) Ωh Ωh

≤ δτ (V m − Vhm )2 Qm |ν m − νhm |2 Qm h + cδ τ h , Ωh Ωh


m m 2 m 3 I3 ≤ δτ (V m − Vhm )2 Qm + cτ |ν − ν | Q + c τ δ h h h m

Ωh

 12

Qm h .

Ωh

We collect the estimates for I1 , I2 , I3 and arrive at the estimate

∇(um+1 − um ) m+1 m · ∇(e − e ) ≤ 3δτ (V m − Vhm )2 Qm A4 = h Qm Ωh Ωh
(m+1)τ

2 2 3 +cδ τ |ν m − νhm |2 Qm + c τ u  + c τ Qm δ t H 2 (Ωh ) δ h h . Ωh

Ωh

mτ

(37) For the third term on the right hand side of (28) we observe 
 m+1 1 u − um A6 = − − um (εm+1 − εm ) t m τ Q Ωh

12
(m+1)τ √ 2 ≤ τ utt L2 (Ωh ) εm+1 − εm L2 (Ωh ) mτ







(m+1)τ

≤ cτ h2 mτ

utt 2L2 (Ωh ) +

(m+1)τ

mτ

ut 2H 2 (Ωh )

(38)

Numerical Approximation of Mean Curvature Flow

and similarly, with the use of (36), 
 m+1 1 m+1 u − um − um (e − em ) A7 = t τ Qm Ωh


m m 2 m m m 2 m 2 ≤ δτ |V −Vh | Qh +cδ τ |ν −νh | Qh +cδ τ Ωh

Ωh

73

(m+1)τ

mτ

utt 2L2 (Ωh ) . (39)

For the last term in (28) we use the fact that H(um ) and V (um ) are bounded uniformly and that |Ωh \ Ω| ≤ ch2 . Thus
− (H(um ) + V (um ))((em+1 − em ) − (εm+1 − εm )) Ωh \Ω

≤ cem+1 − em L1 (Ωh \Ω) + chεm+1 − εm L2 (Ωh )
(m+1)τ ≤ cem+1 − em L1 (Ωh \Ω) + cτ h2 + ch4 ut 2H 2 (Ωh ) .

(40)

mτ

Note further that
em+1 − em L1 (Ωh \Ω) ≤ ≤

Ωh

≤ cτ

Ωh \Ω








|em+1 − em |2  12  Qm h

Ωh \Ω




|V m −Vhm |2 Qm h +

Ωh

 |em+1 − em |   m  m  m | Qh − Q | + c Qh  12   m |2 + ch2 | Qm − Q h

|ν m −νhm |2 Qm h

 12 


Ωh

2 |ν m −νhm |2 Qm h +ch

 12

,

Ωh

where we used (36) together with the estimate


m 2   |Qm 2 h −Q | m m | Qh − Q | ≤ ≤c |νhm − ν m |2 Qm h . Qm Ωh \Ω Ωh Ωh h So, for arbitrary positive δ,
(H(um ) + V (um ))((em+1 − em ) − (εm+1 − εm )) A8 = − Ω \Ω
h
≤ δτ (V m − Vhm )2 Qm + c τ |ν m − νhm |2 Qm δ h h Ωh




Ωh (m+1)τ

+c(τ 2 + h4 ) mτ

2 2 um t H 2 (Ωh ) + cδ τ h .

(41)

We collect the estimates (28,29,33,34–37,38–41), choose δ in a suitable way, and finally get the following result

74

Klaus Deckelnick and Gerhard Dziuk




1 τ 4
+

(V m − Vhm )2 Qm h Ωh

|ν


m+1

−

Ωh

νhm+1 |2 Qm+1 h

−

|ν m − νhm |2 Qm h Ωh


1 + |(ν m+1 − νhm+1 ) − (ν m − νhm )|2 Qm+1 h 8 Ωh
1 2 m + ((Qm+1 − Qm+1 ) − (Qm − Qm h )) /Qh h 8 Ωh

(m+1)τ 2 2 ≤ cτ |ν m − νhm |2 Qm + c(τ + h ) (ut 2H 2 (Ωh ) + utt 2L2 (Ωh ) ) h Ωh mτ
m+1 3 m 2 +cτ (Qh + Qh ) + cτ h . (42) Ωh

 We know, from Theorem 2.2, that Ωh Qm h is bounded and so from the previous estimate it follows that

|ν m+1 − νhm+1 |2 Qm+1 − |ν m − νhm |2 Qm h h Ωh

Ωh




≤ cτ

|ν

m

−

νhm |2 Qm h




2

(m+1)τ

(ut 2H 2 (Ωh ) + utt 2L2 (Ωh ) )

2

+ c(τ + h )

Ωh 3

mτ

+cτ + cτ h2 .

(43)

Summation over m = 0, . . . , l − 1, with l ≤ M − 1, leads to
T
l l 2 l 2 2 |ν − νh | Qh ≤ c(τ + h ) (ut 2H 2 (Ωh ) + utt 2L2 (Ωh ) ) Ωh

0

2

2

+cτ + ch + cτ

l−1


m=0

l−1


≤ c(τ 2 + h2 ) + cτ

m=0

|ν m − νhm |2 Qm h

Ωh

|ν m − νhm |2 Qm h ,

Ωh




or with

|ν l − νhl |2 Qlh ,

l

E = Ωh

E l ≤ cτ

l−1

E m + c(τ 2 + h2 ).

m=0

The discrete Gronwall Lemma then implies E l ≤ c(τ 2 + h2 ), which is the final error estimate for the approximation of the normal,

(44)

Numerical Approximation of Mean Curvature Flow

75


2 m 2 2 |ν(um ) − ν(um h )| Q(uh ) ≤ c(τ + h ).

sup m=0,...,M

(45)

Ωh

This implies the first statement of the Theorem. The second statement concerning the numerical approximation of the normal velocity then follows from (42) if we use (45), i.e. τ

M −1


m=0

2 2 (V m − Vhm )2 Qm h ≤ c(τ + h ).

(46)

Ωh

Thus the Theorem is proved.

3 Mean Curvature Flow of Level Sets 3.1 Viscosity Solutions The level set approach to mean curvature flow allows the modeling of topological changes during the flow. The interface is written as the zero level set of a scalar function, Γ (t) = {x ∈ Rn | u(x, t) = 0} (t ≥ 0).

(47)

As in Section 2, we derive the differential equation for u from the law of motion. Here, ut ∇u V (u) = − , H(u) = ∇ · , |∇u| |∇u| so that (1) is equivalent to ut − |∇u| ∇ ·

∇u = 0. |∇u|

(48)

This differential equation is degenerate in two ways: it is not defined where the gradient of u vanishes – this is the case at points where the topology of Γ changes – and the equation is not uniformly parabolic. We have to solve the following initial value problem ut −

n 

δij −

i,j=1

uxi uxj  uxi xj = 0 |∇u|2 u(·, 0) = u0

in Rn × (0, ∞), in Rn .

(49) (50)

The initial interface Γ0 = {x ∈ Rn | u0 (x) = 0} is given by a continuous function u0 : Rn → R. If u : Rn × [0, ∞) → R is the unique viscosity solution of (48), then we call (47) the generalized solution of the mean curvature flow problem. As Γ (t) exists for all times, it provides a notion of solution

76

Klaus Deckelnick and Gerhard Dziuk

beyond singularities. For this reason, the level set approach has also become very important in the numerical approximation of mean curvature flow and related problems. We shall show how to use the results from the previous chapter to solve (49), (50) numerically in a reliable way. We briefly repeat the analytical framework. Definition 3.1. A function u ∈ C 0 (Rn × [0, ∞)) is called a viscosity subsolution of (49) provided that for each φ ∈ C ∞ (Rn+1 ), if u − φ has a local maximum at (x0 , t0 ) ∈ Rn × (0, ∞), then φt − φt −

n 

δij −

i,j=1 n

φxi φxj  φxi xj ≤ 0 at (x0 , t0 ), if ∇φ(x0 , t0 ) = 0, |∇φ|2



δij −pi pj φxi xj ≤ 0 at (x0 , t0 ) for some |p| ≤ 1, if ∇φ(x0 , t0 ) = 0.

i,j=1

(51)

A viscosity supersolution is defined analogously: maximum is replaced by minimum and
by . A viscosity solution of (49) is a function u ∈ C 0 (Rn × [0, ∞)) that is both a subsolution and a supersolution. We assume that the initial function u0 is smooth. The following existence and uniqueness theorem is a special case of results proved independently by Evans, Spruck [10] and Chen, Giga, Goto [1]. Theorem 3.1. Assume u0 : Rn → R with u0 (x) = 1 for |x| ≥ S for some S > 0. Then there exists a unique viscosity solution of (49), (50), such that u(x, t) = 1 for |x| + t ≥ R, for some R > 0 depending only on S. The authors also proved that the sets Γ (t) are independent of the particular choice of u0 which has Γ0 as its zero level set, so that the generalized evolution (Γ (t))t≥0 is well defined for a given Γ0 . The level set solution has been investigated further in several papers, among which we mention in particular [11], [12], [13] and [20]. 3.2 Regularization Evans and Spruck proved in [10] that the solutions uε of  ∇uε = 0 in Rn × (0, ∞), uεt − ε2 + |∇uε |2 ∇ ·  2 ε + |∇uε |2 uε (·, 0) = u0 in Rn

(52) (53)

converge locally uniformly for ε → 0 to the unique viscosity solution u of (49), (50). We shall use this regularised form of (48) as a basis for the computation of the viscosity solution. Then it is important to know the asymptotic error between the viscosity solution and the solution of the regularised problem quantitatively for  → 0. [2] contains the proof of the following theorem together with several a priori estimates and their dependence on .

Numerical Approximation of Mean Curvature Flow

77

Theorem 3.2. For every α ∈ (0, 12 ), 0 < T < ∞, there is a constant c = c(u0 , T, α) such that u − uε L∞ (Rn ×(0,T )) ≤ cεα for all ε > 0. If one wants to compute approximations to the viscosity solution u of (49), (50) then, according to Theorem 3.2, it is sufficient to solve the regularised problem (52), (53). But one has to restrict the computations to some finite domain. This means that one has to solve the problem  ∇uε uεt − ε2 + |∇uε |2 ∇ ·  = 0 in Ω × (0, T ), (54) 2 ε + |∇uε |2 uε (·, 0) = u0 on ∂Ω × (0, T ) ∪ Ω × {0}, (55) on a domain Ω which is large enough. Corollary 3.1. Let Ω = BS˜ (0) with S˜ > R = R(S), where R is the radius from Theorem 3.1. Then for every α ∈ (0, 12 ), 0 < T < ∞ there is a constant c = c(u0 , T, α) such that u − uε L∞ (Ω×(0,T )) ≤ cεα

(56)

for all ε > 0. Proof. The functions uε and uε are solutions of the same parabolic differential equation in the domain Ω with the same initial data. The comparison theorem then implies that uε − uε L∞ (Ω×(0,T )) ≤ uε − uε L∞ (∂Ω×(0,T )) . On ∂Ω on has that uε = 1 and u = 1. This means that one can continue the estimate by ≤ uε − uL∞ (∂Ω×(0,T )) . But then the result follows from Theorem 3.2. We are now in the position to use the results which were obtained for the mean curvature flow of graphs as a tool for the computation of the viscosity solution to the mean curvature flow of level sets. If we scale uε , (57) U := ε then U is a solution of the mean curvature flow problem for graphs (see (4))  ∇U = 0 in Ω × (0, T ), Ut − 1 + |∇U |2 ∇ ·  1 + |∇U |2 u0 U= on ∂Ω × (0, T ) ∪ Ω × {0}. (58) ε This is a theoretical observation and implies that we can apply the results concerning the mean curvature flow of graphs to the mean curvature flow of level sets. But computationally we shall not use (58), but directly the unscaled version for uε .

78

Klaus Deckelnick and Gerhard Dziuk

3.3 Convergence of the numerical scheme for the level set problem The semi-discrete scheme In the following we use the abbreviations νε (v) =

 (∇v, −ε) vt , Qε (v) = ε2 + |∇v|2 , Vε (v) = − . Qε (v) Qε (v)

Our results for the mean curvature flow of a graph can directly be transformed into a convergence result for the regularised level set problem. Theorem 3.3. Let uε be the solution of (54), (55) and uεh be the solution of the semi-discrete problem uεh (·, t) ∈ Xh with uεh (·, t) − Ih u0 ∈ Xh0 , uεh (·, 0) = uh0 = Ih u0 ,

uεht φh ∇uεh · ∇φh + =0 (59) Q (u ) Qε (uεh ) ε εh Ωh Ωh for all t ∈ (0, T ) and all discrete test functions φh ∈ Xh0 . Then





T

(Vε (uε ) − Vε (uεh ))2 Qε (uεh ) ≤ cε h2 , Ω∩Ωh

0




2

|νε (uε ) − νε (uεh )| Qε (uεh ) ≤ cε h2 .

sup (0,T )

Ω∩Ωh

The proof is based on the scaling argument, which we shall use for the fully discrete scheme below and therefore we omit it here. For a detailed discussion of the dependency on ε see [2]. In two space dimensions we can prove that the computed solutions uεh converge in L∞ to the viscosity solution. The proof is contained in [7]. Theorem 3.4. Let u be the viscosity solution of (49), (50) and let uεh be the solution of the problem 59. with Ω ⊂ R2 as in Corollary 3.1. Then there exist a function h = h(ε) → 0, ε → 0 such that lim u − uεh(ε) L∞ (Ω×(0,T )) = 0.

ε→0

The fully discrete scheme A similar transformation of the graph problem into the level set problem is available for the fully discrete scheme. Let us formulate the algorithm for this case. It can be implemented into an available solver for linear parabolic problems easily. Algorithm 3 Let u0εh = Ih u0 . For m = 0, . . . M − 1 compute um+1 ∈ Xh εh such that uεh − u0εh ∈ Xh0 and for every φh ∈ Xh0 1 τ


Ωh

um+1 εh φh + Q(um εh )


Ωh

∇um+1 · ∇φh 1 εh = m Q(uεh ) τ


Ωh

um εh φh . Q(um εh )

(60)

Numerical Approximation of Mean Curvature Flow

79

We now have the following convergence theorem for the fully discrete regularised level set problem. The required regularity of the continuous solution is available because the domain is a ball and thus mean convex. Theorem 3.5. Let uε be the solution of (54), (55) and let um εh , (m = 1, . . . , M ) be the solution of (60). Then there exists a τ0 > 0 such that, for all 0 < τ ≤ τ0 , M


m 2 m 2 2 τ (Vε (um (61) ε ) − Vεh ) Qε (uεh ) ≤ cε (τ + h ), Ω∩Ωh

m=0




m 2 m 2 2 |νε (um ε ) − νε (uεh )| Qε (uεh ) ≤ cε (τ + h ),

sup m=0,...,M

(62)

Ω∩Ωh

m m = −(um+1 − um with M = [ Tτ ]. Vεh εh )/(τ Qε (uεh )) is the regularised discrete εh normal velocity.

Proof. We scale as in (57). Then U is a solution of the equation for the mean curvature flow of a graph (4) with initial and boundary data U0 = u0 /ε. Let m Uhm = um εh /ε be the scaled solution of the fully discrete scheme (60). Uh then is the solution of the fully discrete scheme (13) for graphs and we can apply Theorem 2.4 and get the error estimates (18), (19). The constants on the right hand side of these estimates now depend on ε because the discrete boundary and initial data depend on this parameter. We scale back the error estimates, observe that m m ν(U ) = νε (uε ), V (U ) = Vε (uε ), ν(Uhm ) = νε (um εh ), Vh = Vεh

and ε Q(U ) = Qε (uε ), ε Q(Uhm ) = Qε (um εh ), so the Theorem is proved. We now dovetail the results from Corollary 3.1 and Theorem 3.5 to get a convergence result for viscosity solutions of the level set problem in R2 . The gradient of the discrete solution is bounded. The gradient of the continuous solution is bounded independently of ε, since the domain is a ball. Let S ∈ Th be a simplex of the triangulation. Then

m m 2 Q(uεh ) ≤ |Q(um ε ) − Q(uεh )| + ch S S
m m 2 ≤ c |ν(um ε ) − ν(uεh )|Q(uεh ) + ch S

 12 



|ν(um ε )

≤c

−

2 m ν(um εh )| Q(uεh )

 12

Q(um εh )

+ ch2


c m 2 m 2 ≤δ + |ν(um ε ) − ν(uεh )| Q(uεh ) + ch δ S S
cε 2 (τ + h2 ) + ch2 , ≤δ Q(um εh ) + δ S


S

Q(um εh )

S

80

Klaus Deckelnick and Gerhard Dziuk

and this implies that


2 2 Q(um εh ) ≤ cε (τ + h ). S

But then sup Q(um εh ) ≤ cε (

τ2 + 1), h2

sup Q(um εh ) ≤ cε (

τ2 + 1). h2

S

and so Ωh

From (27) we then can conclude that m 2 m m 2 m m m m 2 |∇(um ε − uεh )| = |ν(uε ) − ν(uεh )| Q(uεh )Q(uε ) + (Q(uε ) − Q(uεh )) m 2 m m ≤ c|ν(um ε ) − ν(uεh )| Q(uεh )(1 + Q(uεh )) 2 τ m 2 m ≤ cε |ν(um ε ) − ν(uεh )| Q(uεh )(1 + 2 ). h

With the error estimate from Theorem 3.5 we get m ∇(um ε − uεh )L2 (Ωh ) ≤ cε (τ + h)(1 +

τ ) ≤ cε (h + τ ) ≤ cε (h + τ ), h

under the assumption that τ ≤ ch. With the well known estimate [22] 1

wh L∞ (Ωh ) ≤ c| log h| 2 ∇wh L2 (Ωh ) , for functions wh ∈ Xh0 (in R2 ), we can conclude that m m m m m um ε − uεh L∞ (Ωh ) ≤ uε − Ih uε L∞ (Ωh ) + Ih uε − uεh L∞ (Ωh ) 1

m ≤ cε h2 + c| log h| 2 ∇(Ih um ε − uεh )L2 (Ωh ) 1

≤ cε h2 + c| log h| 2  m m m × ∇(Ih um ε − uε )L2 (Ωh ) + ∇(uε − uhε )L2 (Ωh ) 1

≤ cε h2 + cε | log h| 2 (h + τ ) 1

≤ cε | log h| 2 (h + τ ). This estimate, together with the estimate from Corollary 3.1, implies the following bound for the difference between the viscosity solution u and the fully discrete regularised solution um εh m m m m um − um εh L∞ (Ωh ) ≤ u − uε L∞ (Ωh ) + uε − uεh L∞ (Ωh ) 1

≤ cεα + cε | log h| 2 (h + τ ). This is true for arbitrary fixed α ∈ (0, 12 ). Now for ε → 0 we can find h = h(ε) → 0 and τ = τ (ε) → 0 such that with the piecewise linear interpolant uεhτ of um εh with respect to time we have proved the final Theorem.

Numerical Approximation of Mean Curvature Flow

81

Theorem 3.6. Let u be the viscosity solution from Theorem 3.1 and let Ω be the domain from Corollary 3.1. Denote by uεhτ the time interpolated solution of the fully discrete scheme (60). Then there exist functions h = h(ε) → 0 and τ = τ (ε) → 0 for ε → 0 such that u − uεh(ε)τ (ε) L∞ (Ωh ×(0,T )) → 0. 3.4 Numerical Tests for the Level Set Algorithm The regular case. We first test the algorithm for a regular case, i.e. for a case in which the gradient of the solution to (48) does not vanish inside the domain Ω. In order to be able to compute the error between continuous and fully discrete solution we use the exact solution for a shrinking circle, u(x, t) = 8 − 2t − |x|2 , on the domain Ω = {x ∈ R2 | 1 < |x| < 3} for the time interval (0, T ) with T = 2. We use the exact solution as boundary data. Since the gradient of u is nonzero on Ω, we choose the regularization parameter ε = 0. In Table 1 we show the results. For each refinement level of the globally refined grid we provide the grid parameter h, the error of the normal,
 1 2 m 2 sup |ν(u(·, mτ )) − ν(um , Eν = εh )| |∇uεh | m∈{0,...,M }

Ωh

the error for the normal velocity, −1

 M EV = τ m=0




12

2 m (V (u(·, mτ )) − V (um εh )) |∇uεh |

,

Ωh

where we have set ν(w) =

∇w , |∇w|

V (w) = −

wt , |∇w|

for a generic function w, and the error for the approximation of u by uεhτ in the norms EL∞ (L2 ) = u − uεhτ L∞ ((0,T ),L2 (Ωh )) and EL2 (H 1 ) = u − uεhτ L2 ((0,T ),H 1 (Ωh )) , where we denote by uεhτ the piecewise linear interpolation of um εh (m = 0, . . . , M ) with respect to time. Between two spatial discretization levels h1 and h2 we compute the experimental order of convergence

82

Klaus Deckelnick and Gerhard Dziuk

eoc(h1 , h2 ) = log

h1 E(h1 ) ! log , E(h2 ) h2

for the errors E(h1 ) and E(h2 ), and for each of the error norms Eν , EV , EL∞ (L2 ) and EL2 (H 1 ) . For the time step size we have used τ = 0.01h. Obviously the results of the Theorem are reproduced in practical computations. We also observe second order convergence for the L2 -norm and linear convergence for the H 1 -norm. Table 1. The regular case: errors and experimental orders of convergence for the test problem

lev.

h

Eν

eoc

EV

eoc

EL∞ (L2 )

eoc

EL2 (H 1 )

eoc

2

2.1248

1.2140

-

1.3400

-

3.1444

-

4.7184

-

4

1.1798

0.5748

1.27

0.8245

0.83

0.9004

2.13

2.5803

1.03

6

0.6161

0.2730

1.15

0.4063

1.09

0.2234

2.15

1.2476

1.12

8

0.3142

0.1318

1.08

0.1962

1.08

0.05678

2.03

0.6118

1.06

10

0.1585

0.06454

1.04

0.09469

1.07

0.01426

2.02

0.3017

1.03

Mildly singular case. As a first test for the singular case we use the same exact solution as in the previous section, but now we choose the domain to be Ω = {x ∈ R2 | |x| < 3}. There is exactly one singular point within the domain now, namely x = 0. The results are shown in Table 2. Table 2. The mildly singular case: errors and experimental orders of convergence for the test problem with the choice ε = h

lev.

h

Eν

eoc

EV

eoc

EL∞ (L2 )

eoc

EL2 (H 1 )

eoc

2

3.0000

2.8489

-

4.6263

-

15.7814

-

7.5245

-

4

1.7132

1.4571

1.19

3.9263

0.29

6.3723

1.62

4.2222

1.03

6

0.9137

0.6733

1.24

2.9242

0.47

1.8536

1.96

2.0767

1.13

8

0.4713

0.3015

1.21

2.1304

0.48

0.4845

2.03

0.8617

1.33

10

0.2393

0.1382

1.15

1.5250

0.49

0.1219

2.04

0.3822

1.20

Numerical Approximation of Mean Curvature Flow

83

We have chosen ε = h and τ = 0.01h and we have set ν(u) := 0 where ∇u = 0. The results for similar parameters but with ε = h2 are shown in Table 3. We observe convergence for ε = h but the order of convergence for the normal velocities may break down for the choice ε = h2 . Recall, that we did not prove convergence of the discrete solution to the viscosity solution in stronger norms than the L∞ (Ωh × (0, T ))-norm. Thus we only expect convergence for the L∞ ((0, T ), L2 (Ω))-norm. But the fact that the set of singular points is very small in this case may be responsible for convergence of the computed solution to the viscosity solution for higher norms in this case. Table 3. The mildly singular case: errors and experimental orders of convergence for the test problem with the choice ε = h2

lev.

h

Eν

eoc

EV

eoc

EL∞ (L2 )

eoc

EL2 (H 1 )

eoc

2

3.0000

2.7983

-

4.9629

-

16.1827

-

7.8973

-

4

1.7132

1.4510

1.17

4.7450

0.08

8.9268

1.06

6.0774

0.47

6

0.9137

0.6740

1.22

2.8106

0.83

1.6596

2.68

2.0154

1.76

8

0.4713

0.3030

1.21

1.6609

0.80

0.2517

2.85

0.7483

1.50

10

0.2393

0.1389

1.15

1.4901

0.16

0.05850

2.15

0.4523

0.74

The highly singular case. As a computational example for a highly singular case with a large singular set {x ∈ Ω| ∇u(x) = 0}, we take a mathematically trivial viscosity solution of the mean curvature flow for level sets. But it is numerically very interesting. We choose the piecewise linear function   x1 + 1 : x1 ∈ [−3, −1] 0 : x1 ∈ [−1, 1] u0 (x1 , x2 ) =  x1 − 1 : x1 ∈ [1, 3]. The level lines of u0 are straight lines, i.e., their curvature is zero. Therefore they do not move by mean curvature flow. The viscosity solution u is given by u = u0 . Nevertheless the singular set is the strip {x ∈ R2 | |x| < 3, x1 ∈ (−1, 1)} and so is a large set. We only expect convergence for the L∞ ((0, T ), L2 (Ω))-norm. √ The results are shown in Table 4. The parameters are τ = 0.01h, ε = 0.01 h. The dumb-bell problem. In Figure 4 we show the results of the computation of the mean curvature flow for a dumb-bell type surface. The computation uses the numerical method from [8] and naturally breaks down when the singularity appears. We also used the numerical methods for the computation of viscosity solutions for the level set formulation of mean curvature flow from

84

Klaus Deckelnick and Gerhard Dziuk

Table 4. The highly singular case: errors and √ experimental orders of convergence for the test problem with the choice ε = 0.01 h

lev.

h

Eν

eoc

EV

eoc

EL∞ (L2 )

eoc

EL2 (H 1 )

eoc

2

3.0000

2.7027

-

0.4385

-

1.7160

-

2.5481

-

4

1.7132

1.1996

1.45

0.3454

0.43

0.2753

3.27

0.7667

2.14

6

0.9137

1.2003

0.00

0.3553

-0.05

0.1955

0.54

0.7988

-0.07

8

0.4713

0.9483

0.36

0.3043

0.23

0.1024

1.16

0.7592

0.08

10

0.2393

0.6688

0.52

0.2602

0.23

0.04739

1.14

0.5688

0.43

this paper to compute the evolution across the singularity for the same initial surface. We have used the oriented distance function of the initial dumb-bell as initial value for the level set problem. We kept the boundary data fixed. The problem is axially symmetric and so we have solved  the corresponding two dimensional problem in the x1 -r-plane, where r = x22 + x23 . The computational domain was Ω = (−2, 2) × (−2, 2). The differential equation for the axially symmetric problem is similar to the twodimensional problem except for a weight r in the integrals. Figure 2 shows level lines of the initial function and level lines of the solution at the time when the surface breaks apart. Figure 3 shows the results for several time steps.

Fig. 2. Level lines of the initial values (left) and of the solution at the 325-th time step for the axially symmetric dumb-bell

Numerical Approximation of Mean Curvature Flow

85

Fig. 3. The 0-level line for the axially symmetric dumb-bell problem at time steps 0, 200, 314, 325, 400 and 1000

86

Klaus Deckelnick and Gerhard Dziuk

Fig. 4. Formation of a singularity in mean curvature flow

References [1]

Chen, Y–G., Giga, Y. & Goto, S. Uniqueness and existence of viscosity solutions of generalized mean curvature flow equations. J. Diff. Geom 33 (1991), 749–786

Numerical Approximation of Mean Curvature Flow [2] [3] [4] [5]

[6] [7]

[8] [9]

[10] [11] [12] [13] [14] [15] [16]

[17]

[18]

[19] [20] [21] [22]

87

Deckelnick, K. Error analysis for a difference scheme approximating mean curvature flow. Interfaces and Free Boundaries 2 (2000), 117–142 Deckelnick, K. & Dziuk, G. Convergence of a finite element method for non–parametric mean curvature flow. Numer. Math. 72 (1995), 197–222 Deckelnick, K. & Dziuk, G. Discrete anisotropic curvature flow of graphs. M2AN Math. Model. Numer. Anal. 33 (1999), 1203–1222 Deckelnick, K. & Dziuk, G. Error estimates for a semi implicit fully discrete finite element scheme for the mean curvature flow of graphs. Interfaces and Free Boundaries 2 (2000), 341–359 Deckelnick, K. & Dziuk, G. A fully discrete numerical scheme for weighted mean curvature flow. Numer. Math. 91 (2002), 423–452 Deckelnick, K. & Dziuk, G. Convergence of numerical schemes for the approximation of level set solutions to mean curvature flow. Preprint Mathematische Fakult¨ at Freiburg (2000), 1–17 Dziuk, G. An algorithm for evolutionary surfaces. Numer. Math. 58 (1991), 603–611 Dziuk, G. Numerical schemes for the mean curvature flow of graphs. In P. Argoul, M. Fr´emond, Q. S. Nguyen (Eds.): IUTAM Symposium on Variations of Domains and Free-Boundary Problems in Solid Mechanics. 63–70, Kluwer Academic Publishers, Dordrecht-Boston-London 1999 Evans, L.C. & Spruck, J. Motion of level sets by mean curvature I. J. Diff. Geom. 33 (1991), 636–681 Evans, L.C. & Spruck, J. Motion of level sets by mean curvature II. Trans. Amer. Math. Soc. 330 (1992), 321–332 Evans, L.C. & Spruck, J. Motion of level sets by mean curvature III. J. Geom. Anal. 2 (1992), 121–150 Evans, L.C. & Spruck, J. Motion of level sets by mean curvature IV. J. Geom. Anal. 5 (1995), 77–114 Fried, M. Niveaufl¨ achen zur Berechnung zweidimensionaler Dendrite. Dissertation Freiburg 1999 Huisken, G. Non–parametric mean curvature evolution with boundary conditions. J. Differential Equations 77 (1989), 369–378 Lieberman, G. A. The first initial–boundary value problem for quasilinear second order parabolic equations Ann. Sci. Norm. Sup. Pisa Ser. IV 8 (1986), 347–387 Oliker, V.I. & Uraltseva, N.N. Evolution of nonparametric surfaces with speed depending on curvature II. The mean curvature case. Comm. Pure Appl. Math. 46 (1993), 97–135 Osher, S. & Sethian, J.A. Fronts propagating with curvature dependent speed: Algorithms based on Hamilton–Jacobi formulations. J. Comp. Phys. 79 (1988), 12–49 Sethian, J.A. Level set methods. Cambridge Monographs on Applied and Computational Mathematics 3, Cambridge University Press (1996) Soner, H.M. Motion of a set by the curvature of its boundary. J. Differ. Equations 101 (1993), 313–372 Walkington, N.J. Algorithms for computing motion by mean curvature. SIAM J. Numer. Anal. 33 (1996), 2215–2238 Yserentant, H. On the multi–level splitting of finite element spaces. Numer. Math. 49 (1986), 379–412

Reaction-Diffusion Systems Arising in Biological and Chemical Systems: Application of Singular Limit Procedures Masayasu Mimura Department of Mathematical and Life Sciences Institute of Nonlinear Sciences and Applied Mathematics Graduate School of Science, Hiroshima University, Japan [email protected]

Contents 1. 2. 3. 4. 5. 6.

What is diffusion? Paradox of diffusion Diffusive patterns and waves-bistable RD equations Resource-consumer RD systems Competition-diffusion systems and singular limits procedures Chemotactic patterns - aggregation and colonies

1 What is diffusion? Diffusion enhances spatial homogeneization. Nonlinear phenomena related to diffusion appear in chemistry, biology and other scientific fields. In order to theoretically understand these phenomena, many discrete and continuous diffusion models have been proposed. 1.1 Discrete diffusion models Two cells model We consider the simple situation where two cells C1 and C2 are connected by a diffusive membrane and u1 (t) and u2 (t) be the concentrations of a single chemical species at time t in C1 and C2 . Assuming that each species flows from one cell to the other, in proportional to the difference of their concentrations, the time evolution of u1 (t) and u2 (t) is simply given by u1 (t) = d[u2 (t) − u1 (t)] u2 (t) = d[u1 (t) − u2 (t)]

,

t > 0,

(1)

where  indicates the differentiation with respect to time and d is the diffusion coefficient of the membrane. By using the property u1 (t) + u2 (t) = u1 (0) + u2 (0), one can easily obtain

L. Ambrosio et al.: LNM 1812, P. Colli and J.F. Rodrigues (Eds.), pp. 89–121, 2003. c Springer-Verlag Berlin Heidelberg 2003


90

Masayasu Mimura

lim u1 (t) = lim u2 (t) =

t→∞

t→∞

u1 (0) + u2 (0) , 2

(2)

that is, the concentration of the species in the two cells becomes equal. This clearly implies that (1) possesses the effect of diffusion. N cells model Let u1 (t), u2 (t),. . . , uN (t) be the concentrations of a single chemical species in N cells C1 , C2 ,. . . , CN , which are one-dimensionally connected by diffusive membranes. A diffusion model corresponding to (1) is u1 (t) = d[u2 (t) − u1 (t)] = d∇ + u1 (t) u2 (t) = d[u1 (t) − u2 (t)] + d[u3 (t) − u2 (t)] = d∇ + ∇− u2 (t) = d∆+− u2 (t) ... ui (t) = d[ui−1 (t) − ui (t)] + d[ui+1 (t) − ui (t)] = d∆+− ui (t) ... uN −1 (t) uN (t) =

= d[uN −2 (t) − uN −1 (t)] + d[uN (t) − uN −1 (t)] = d∆+− uN −1 (t) d[uN −1 (t) − uN (t)] = −d∇ − uN (t) (3)

for t > 0. Similarly to (2), we can also show lim ui (t) =

t→∞

u1 (0) + u2 (0) + · · · + uN (0) N

(i = 1, 2, . . . , N ),

(4)

that is, ui (t) (i = 1, 2, . . . , N ) tends to the average of u1 (0), u2 (0), . . . uN (0). System (3) can be easily extended to general two-dimensional discrete model for uij (t) in N 2 cells Cij (i, j = 1, 2, . . . , N ). 1.2 Continuous models We point out that (3) can be naturally represented in the continuous version. Let u(t, x) be the concentration or the density of a single species in isotropic diffusive medium at time t and position x. Then, it is described by ut = d∆u,

t > 0, x ∈ Ω,

(5)

where d is the diffusion rate, ∆ is the Laplacian operator in Rn . Consider (5) in a bounded domain Ω in Rn under the following initial and boundary conditions u(0, x) = a(x), ∂u = 0, ∂ν

x ∈ Ω,

(6)

t > 0, x ∈ ∂Ω,

(7)

where ν is the outerward normal vector on the boundary ∂Ω. One knows

Reaction-Diffusion Systems

lim u(t, x) =< a >,

91

(8)

t→∞

where < f > means the spatial average of f (x) in Ω. It turns out that the concentration becomes homogeneous in space, which is obviously a continuous version of (4). Furthermore, (5) is extended to ut = d∆u + f (u) ,

t > 0, x ∈ Ω,

(9)

where f (u) is a reaction term (9) is simply called a scalar reaction-diffusion (RD) equation. In the fields of chemistry, ecollogy and biology, a lot of models described by the form (9) have been proposed so far. For the asymptotic behavior of solutions of (9) with (6) and (7), the following results are already known: (i) w-limit set consists of equilibrium solutions only [1]; (ii) when the domain Ω is convex, any non-constant equilibrium solutions are unstable, even if they exist [2]; (iii) there are suitable non-convex domain Ω and f (u) such that there are stable non-constant equilibrium solutions [3]. These properties indicate that if the domain Ω is convex, any bounded solution generically approach one of the stable constant equilibrium solutions which are given by stable critical points of the diffusionless equation of (9) ut = f (u) ,

t > 0.

(10)

This result seems to be obvious, because we already know that diffusion enhances spatial homogeneization. Therefore, it was long believed that RD equations could not generate any spatial patterns and waves and therefore these were not interested from pattern formation viewpoints.

2 Paradox of diffusion In 1952, two surprising paradoxical evidences of diffusion were demonstrated by using mathematical models. The first contributor is a mathemacian, A. Turing, who stated that diffusion enhances spatial inhomogeneities [4]. In order to explain this paradox, he proposed a discrete two cells model, in which two different types of chemical species, one activator and its inhibitor, are included. Let Ai (t) and Ii (t) be respectively the concentrations of the activator and its inhibitor at time t in the i-th cell (i = 1, 2). The resulting system for (ui (t), vi (t)) (i = 1, 2) is described by the following four-dimensional ODEs A1 = dA [A2 − A1 ] + f (A1 , I1 ) I1 = dI [I2 − I1 ] + g(A1 , I1 ) A2 = dA [A1 − A2 ] + f (A2 , I2 ) I2 = dI [I1 − I2 ] + g(A2 , I2 )

,

t > 0,

(11)

92

Masayasu Mimura

where dA and dI are respectively the diffusion coefficients of the activator and its inhibitor. f and g describe the interaction of the activator and its inhibitor. The interaction is given as follows (i) Activator grows in an autocatalytic way; (ii) Activator produces inhibito, by which its growth is inhibited; (iii) Inhibitor decays in a constant rate. When A1 = A2 = A and I1 = I2 = I, (11) reduces to A = f (A, I) I  = g(A, I)

,

t > 0.

(12)

Suppose that there is a critical point, say (A, I), of (12). Then (A1 , I1 ; A2 , I2 ) = (A, I, ; A, I)

is a critical point of (11).

We call it a homogeneous critical point, because the states in two cells C1 and C2 are equal. Assume that (A, I) is stable in (12). Turing claimed that the critical point (A, I; A, I) is not necessarily stable in (11). In fact, if dI is suitably larger than dA , the critical point is unstable in (11), in other words, the difference of A1 (t) and A2 (t) (and also I1 (t) and I2 (t)) increases with time. This is called diffusion-induced instability. In order to demonstrate this instability explicitly, we propose a linearized system of (12) with a critical point (A, I) = (0, 0) for simplicity, A = 5A − 6I I  = 6A − 7I

,

t > 0.

(13)

It is obvious to see that (0, 0) is asymptotically stable, that is, lim (A(t), I(t)) = (0, 0).

t→∞

We now consider the two cell systems for (A1 , I1 ; A2 , I2 ) corresponding to (11) A1 = dA [A2 − A1 ] + 5A1 − 6I1 I1 = dI [I2 − I1 ] + 6A1 − 7I1 A2 = dA [A1 − A2 ] + 5A2 − 6I2

,

t > 0.

(14)

I2 = dI [I1 − I2 ] + 6A2 − 7I2 Of course, (A1 , I1 ; A2 , I2 ) = (0, 0; 0, 0) is the homogeneous critical point which is independent of the diffusion rates dA and dI . In order to study the stability of this critical point, we introduce four new variables P, Q, R and S by A1 + A2 = P, I1 + I2 = Q,

A1 − A2 = R, I1 − I2 = S.

(15)

Reaction-Diffusion Systems

93

Then (14) is rewritten as P  = P − 6Q Q = 6P − 7Q

(16)

and R = 5R − 6S − 2dA R S  = 6R − 7S − 2dI S

.

(17)

One find that (16), (17) are decoupled. Since (16) is the same as (13), we see that the critical point (P, Q) = (0, 0) is asymptotically stable, that is, lim (A1 (t) + A2 (t), I1 (t) + I2 (t)) = (0, 0).

t→∞

(18)

We next consider (17). The stability analysis indicates the following: if dA and dI satisfy (5 − 2dA )(7 + 2dI ) < 36, the critical point (R, S) = (0, 0) is asymptotically stable, while if (5 − 2dA )(7 + 2dI ) > 36, it is unstable in (dA , dI ) - plane, as shown in Fig. 1.

(a)

(b)

Fig. 1. Bifurcation curve in (dA , dI )-plane

Consequently, if (R, S) = (0, 0) is asymptotically stable, we infer lim (A1 (t) − A2 (t), I1 (t) − I2 (t)) = (0, 0),

t→∞

(19)

so that, by (18), lim A1 (t) = lim A2 (t) = 0

t→∞

t→∞

and

lim I1 (t) = lim I2 (t) = 0,

t→∞

t→∞

that is, the homogeneous critical point (0, 0; 0, 0) is asymptotically stable. On the other hand, if (R, S) = (0, 0) is unstable, even if the difference of A1 (0) and A2 (0) or I1 (0) and I2 (0) is small, the differences of A1 (t) and

94

Masayasu Mimura

A2 (t) or I1 (t) and I2 (t) are exponentially increasing. In fact, suppose dA is arbitrarily fixed in order to satisfy 0 < dA < 2/5. By Fig. 1, one finds that, if dI is small, (0, 0; 0, 0) is asymptotically stable; however, if dI is increasing, it becomes unstable. This is the diffusion-induced instabilty which Turing originally stated in 1952. The idea of the above is naturally extended to an N-cells model and furthermore, to the following two component RD equations ut = du ∆u + f (u, v) vt = dv ∆v + g(u, v)

,

t > 0, x ∈ Ω ⊂ Rn ,

(20)

in a bounded domain Ω, where du and dv are the diffusion coefficients of u and v, respectively. Under the zero flux boundary conditions, numerical calculations reveal that RD systems of the form (20) generate a surprisingly variety of spatial and/or spatio-temporal patterns, when f and g fall into the framework of activator-inhibitor interaction. The second contributors are two neurophysiologists, A. L. Hodgkin and A. F. Huxley who investigated the mechanism of impulses with constant shape and velocity propagating along nerve fiber. They proposed a RD model to describe the propagation of impulses along the fiber [5]. The model is given by the coupling of a single RD equation and three ODEs. Numerical simulation showed that their model generated a traveling pulse which represents propagating impulse. This is also another paradoxical evidence of diffusion in a sense that diffusion generate a localized wave, if it is coupled with some reaction process. These two evidences arising in RD equations seem to contradict the result stated in the previous section. However, we should remark that the equations which generate such patterns and waves are not single equations but systems of the following form Ut = D∆U + F (U ), (21) where U (t, x) = (u1 , u2 , . . . , uN )(t, x), D is an N × N diagonal matrix with non-negative elements {di }(i = 1, 2, . . . , N ). Moreover, F (U ) = (f1 (U ), f2 (U ), . . . , fN (U )) describes suitable reaction terms which represents the interaction among U. In the following section, as an example, we will introduce bistable RD equations and demonstrate how the interplay of diffusion and reaction generates complex spatio-temporal patterns which can not be expected by the diffusionless system of (21).

3 Diffusive patterns and waves - bistable RD equations In general, RD equations of the form (21) consist of two well studies equations. One is the heat equation to describe “diffusion process” ut = d∆u

(22)

Reaction-Diffusion Systems

95

and the other is a system of ODEs to describe “reaction process” Ut = F (U ).

(23)

Then, a naive question arises: can qualitative properties of solutions to RD equations (21) be known by the information of the diffusionless equations (23)? In the next section, we will show that this can not be necessarily expected. 3.1 Scalar bistable RD equation Let us first consider a scalar RD equation with cubic nonlinearity f (u) = u(1 − u)(u − a) satisfying 0 < a < 1 ut = d∆u + f (u),

t > 0, x ∈ Ω ⊂ Rn ,

(24)

where d is the diffusion coefficient. We simply assume that Ω is a bounded and convex domain. One easily knows that stable equilibrium solutions are spatially constant equilibria u = 0 and u = 1 only, while the rest u = a is unstable. Therefore, we call that (22) is a scalar bistable RD equation. First consider the one dimensional problem of (24) ut = duxx + f (u),

t > 0, x ∈ (−L, L),

(25)

under the zero flux boundary conditions x = ±L and the initial conditions u(0, x) = 1

for

−L < x < 0

u(0, x) = 0

for

0<x
.

(26)

Then, because of bistability, one can expect the occurrence of a single transition layer, connecting two states where u takes nearly 1 at the left hand side and 0 in the right. After that, the layer generally propagates to either the wall x = L or x = −L, as if its shape and velocity are both constant, as in Fig. 2, and eventually it approaches to the wall and then the solution u becomes one of the constant states: either u = 0 or u = 1. This expectation arrives at that the final state would be either u = 0 or u = 1. It turns out that the result is the same as the one of the corresponding ODE ut = f (u),

t > 0.

(27)

Only the difference is the transient behavior of solutions. Here we emphasize that the information on travelling front solutions play an important role on the study of this behavior.

96

Masayasu Mimura

Fig. 2. Layer propagation in (25)-(26)

(i) Travelling front solutions We consider a special solution of the form u(t, x) = u(z) with z = x − ct in −∞ < z < ∞, where c is called the velocity. It satisfies −cu = du + f (u),

−∞ < z < ∞,

(28)

with the boundary conditions u(−∞) = 1 and u(∞) = 0.

(29)

For the √ problem (28), (29), the velocity c is explicitly given by c = c(a) = (1−2a)/ 2 for which there is a unique solution φ(z; a) except for translational free. If 0 < a < 1/2 (resp. 1/2 < a < 1), c > 0 (resp. c < 0), φ(x − c(a)t; a) propagates to the right (resp. left) direction. We therefore call it a travelling front. In particular, for the equi-stable case when a = 1/2, c = 0 so that φ(x; 1/2) is a nonconstant equilibrium solution, which is called a standing front. By this information on travelling velocity, one can expect that if c > 0, the layer in (25), (26) propagates to the right direction so that the final state solutions could be u = 1, while if c < 0, the situation is opposite. The equistable case when c = 0 is exceptional. The standing wave solution indicates that the layer does not move, as if the solution u is an equilibrium solution of (25). However, we already know that any nonconstant equilibrium solutions are unstable so that the layer should move under small perturbations. Thus, the following question naturally arises: “how does the transition layer move?”. Unfortunately, the standing wave solution does not answer to this question. It will be explained later. Except for the equi-stable case, the transient as well as asymptotic behaviors of the one-dimensional case of (24) are inherited to the two-dimensional case where the initial function u(0, x, y) is a suitably large delta-like function. If c > 0 (0 < a < 1/2), u(t, x) grows and expands uniformly so that the final state is u = 1, while if c < 0 (1/2 < a < 1), it shrinks so that the final state is u = 0. The equi-stable case when a = 1/2 is very delicate similarly to the one-dimesional case.

Reaction-Diffusion Systems

97

(ii) Slow diffusion and fast reaction limit In order to know the time evolution of transition layer, we consider the special situation where diffusion is very slow and reaction is very fast so that (24) is rewritten as the following RD equation, with a small parameter ε > 0, ut = ε∆u + ε−1 f (u),

t > 0, (x, y) ∈ Ω ⊂ R2 .

(30)

Because ε is sufficiently small, there generally appear transition sharp layer regions with narrow width in a short time, which seperate two qualitatively different regions where u takes nearly either 1 or 0. Singular perturbation arguments say that such layer regions become interfacial curves, say Γ (t), in the limit ε ↓ 0. For sufficiently small ε > 0, the time evolution of such layer regions is described well approximately by the following evolution equation for the interface Γ (t) V (t) = c(a) − εκ,

t > 0, (x, y) ∈ Γ (t),

(31)

where V (t) is the normal velocity of the interface Γ (t) and κ is the curvature of the interfacial curve [6]. If the initial interface Γ (0) is given by a circle with radius r0 , for instance, then the resulting interface Γ (t) is also given by a circle with radus r(t) which is a solution of r = c(a) − ε/r, r(0) = r0 .

t > 0,

(32)

When c(a) > 0, one ovbiously finds that, if r0 > ε/c(a), the circular interface expands, while when c(a) < 0, it shrinks to zero in finite time. This suggests us that u(t, x) of (30) eventually tends to the constant state u = 1 for the former case, and u = 0 for the latter case. In particular, for the equi-stable case when a = 1/2 or c = 0, (31) reduces to an equation for motion of curvature V (t) = −εκ,

t > 0,

(33)

which has been intensively investigated in the fields of differential geometry and nonlinear PDEs ([7], for instance). For the equi-stable case, we find that the curvature effect generically plays an important role on the dynamics of layers in higher dimensions. A special case is the one dimensional problem where curvature effect is absent, for which an interfacial curve is given by one point, say x = ζ(t), so that the time evolutional equation for ζ(t) is simply given by ζ  (t) = 0,

(34)

which says that the interfacial point does not move for any location in the interval (−L, L). However, if ε is not zero, the correponding layer in (25) has to generically move. It is shown in [8] that its velocity is extremely slow.

98

Masayasu Mimura

In the following, we fix a = 1/2 (equi-stable case) and consider (30) within a rectangular domain Ω = {(x, y) ∈ R2 | − L < x < L, −L < y < L} with zero-flux boundary conditions. Take an initial interfacial curve x = ξ(y) specified by u(0, x, y) = 1 u(0, x, y) = 0

−L ξ(y)

< x < ξ(y), < x < L,

−L < y < L, −L < y < L,

(35)

where x = ξ(y) is given by the line x = 0 with a small deformation. As shown in Fig. 3, numerical simulation indicates that the deformation is instantly decaying and the resulting interfacial curve becomes almost straight and moves extremely slowly in the one-dimension fashion. We thus find that the final state is very simple, that is, either u = 1 or u = 0. This result is not surprising, because the system consists of diffusion and bistable reaction.

Fig. 3. Transversal stability of interfacial curve in (30)

Finally, in higher dimensions with n ≥ 3, (31) can be extended to V (t) = c(a) − ε(n − 1)κ,

t > 0,

(36)

where n is the spatial dimenionality and κ is the mean curvature of the interface. For the property of time evolution of interfaces, the readers should refer to [9, 10] and the references therein. 3.2 Two component systems of bistable RD equations We consider the following two component RD system, with a small parameter ε > 0 ετ ut = ε2 ∆u + f (u) − v vt = d∆u + u − γv

,

t > 0, x ∈ Ω ⊂ R2 ,

(37)

Reaction-Diffusion Systems

99

where τ, ε and γ are positive constants and f (u) = u(1 − u)(u − a) with 0 < a < 1. This is one of typical RD systems which Turing proposed to explain diffusion induced instability, where the unknowns u and v are respectively called the activator and its inhibitor. Suppose that γ is suitably large such that the nulclines f (u) = v and u = γv intersect at three points, say, P, Q and R, as in Fig. 4. If γ tends to infinity, v becomes zero so that the first equation of (37) reduces to the scalar equation of the same type as (24). In this sense, (37) is a natural extension of (24).

Fig. 4. Curves of f (u) = v and u = γv in (u, v)-plane

First, we consider the diffusionless system of (37), that is, ετ ut = f (u) − v vt = u − γv

,

t > 0.

(38)

One finds that the critical points P and R are asymptotically stable, while Q is unstable. That is, (37) is a bistable RD system. Let γ = γ ∗ be the value such that P and R are odd symmetric with respect to Q. This situation implies that P and R are equi-stable (which corresponds to a = 1/2 in (24)). Consider (37) in a rectangluar domain Ω = {(x, y) ∈ R2 | − L < x < L, −L < y < L} with zero-flux boundary conditions. We first note that P and R are stable spatially homogeneous equilibrium solutions of (37). Therefore, the following question arises: “Are there any other asymptotic states of solutions in addition to the spatial constant equilibria P and R?”. In order to answer this question, we fix γ = γ ∗ and also fix ε sufficiently small and a such that 0 < a < 1/2, respectively. We take t as a free parameter. Specify the following initial conditions, which are similar to the scalar case (35), as (u(0, x, y), v(0, x, y)) = P (u(0, x, y), v(0, x, y)) = R

− L < x < ξ(y), −L < y < L, ξ(y) < x < L, −L < y < L.

(39)

Since ε is sufficiently small, singular perturbation arguments say that there appear transition sharp layers in the u-component, while the other v is smooth

100

Masayasu Mimura

[11]. We are now interested in time evolution of layer in the u-component only. For large τ , the initial deformation instantly decays and the resulting layer becomes uniform with respect to the y-direction and it slowly moves to the center line x = 0, where the solution tends to an odd symmetric equilibrium. This can be expected, because the nonlinearity of f and g takes odd-symmetry. On the other hand, for small τ , the situation is drastically changed. Some parts of the curve x = ξ(y) move to the left direction and some parts move to the right direction so that the resulting curve becomes very complicated. Numerical calculation strongly suggests the coexistence of two travelling fronts which propagate to the different directions (for the rigorous proof, see [12]). As a special situation, we take the initial interface such that one part of x = ξ(y) (0 < y < L/2) moves to the left direction and the other part (L/2 < y < L) does to the right. Then the resulting pattern exhibits a steadly rotating spiral with two arms (Fig. 5).

Fig. 5. Spiral wave with two arms arising in (37)

For intermediate values of τ , the resulting pattern is totally different from the aboves. We take the same initial interface as the above. Initial pattern begins to envolve into spiral but it gradually breaks down and changes to a very complicated labyrinthine pattern after large time (Fig. 6). The occurrence of such spiral and labyrinthine patterns can not be expected from the corresponding ODEs (38). This numerical simulations show that, although the interaction of activator and its inhibitor possesses simple bistable mechanism, the interplay of this interaction with diffusion possibly generates unexpected complex spatiotemporal patterns, even if the system looks so simple. It is now clear that the study of qualitative properties of solutions of RD systems is very challenging from not only applied sciences but also purely mathematical viewpoints.

Reaction-Diffusion Systems

101

Fig. 6. From break up of spiral to labyrinthine pattern in (31)

4 Resource-consumer RD systems In this section, we will discuss a class of RD systems which possess consumerresource interaction, which arise in fields of biology and chemistry. 4.1 Bacterial colony model The diversity and complexity of spatio-temporal patterns arising in biological systems often take place through the intertwinning of complex biological factors and environmental conditions. Intensive investigations have shown that the growth of bacterial colonies produces various complex patterns, depending on the species and environmental conditions. For instance, the bacterial species called Bacillus subtilis is known to exhibit qualitatively different twodimensional colony patterns as a result of growth and cell division on the surface of thin agar plates by feeding on nutrients [13], [16]. Colony patterns change drastically when the concentrations of agar and nutrient, say Ca and Cn , are globally varied. They are qualitatively classified into five types, each of which is observed in the regions labeled A−E in the (Ca , Cn−1 )-plane, as in Fig. 7. In the region A (hard agar medium containing poor nutrient), colony patterns exhibit tip-splitting growth with characteristically branched structures (Fig. 8). These patterns are similar to those observed in diffusion limited processes in solidification from a supersaturated solution, solidification from an undercooled liquid, and electro-chemical deposition [14]. Also, it was reported that averages over about 25 samples of colony patterns in experiments yield a fractal dimension of 1.72±0.02. This is indeed in good agreement with the results from two-dimensional DLA particle model. Increasing Cn with high fixed Ca values (corresponding to the change from the region A to B), the branch thickness of the colony increases gradually and colony patterns eventually become Eden-like. For high Cn and low Ca (soft agar medium

102

Masayasu Mimura

with rich nutrient) in the region D, colony patterns drastically change to be spread homogeneously, which look macroscopically like a perfect disk. It is likely that the movement of bacterial cells inside a colony can be described in terms of diffusion, and there is no microscopic branching at all. In the region E, between the regions A and D, there emerge colony patterns clearly reminiscent of the so-called dense-branching morphology (DBM). Though the branching is very dense, the advancing envelope looks characteristically smooth compared with DLA-like colonies in the region A. Finally, in the region C, between the regions B and D, colonies spread and rest alternately, leaving stationary concentric ring-like patterns.

Fig. 7. Phase diagram of colony patterns

Fig. 8. Branched pattern in Region A

The above experimental observations lead to the following naive question Is the diversity of colony patterns observed in experiments caused by different effects or governed by the same underlying mechanisms?

Reaction-Diffusion Systems

103

To theoretically understand the mechanism generating branched colony patterns, a cell-walker discrete model coupled with a diffusion equation for the nutrient concentration was proposed [15]. The numerical results confirm that the model generates patterns similar to the ones observed in the regions A and E. On the other hand, a continuous model of RD equation for growth patterns in the region D was proposed [16]. Let b(t, x) be the population density of bacterial cells at time t and position x. The equation describing b(t, x) is bt = db ∆b + (b0 − µb),

(40)

where db is the diffusion coefficient of bacterial cells, b0 is the initial concentration of nutrients, and µ is the intraspecific competition rate of cells. The parameters are all positive constants. The type of (40) is called the Fisher equation in mathematical genetics. If the initial function b(0, x) is a point-like distribution, the solution b(t, x) spreads like an expanding disk. It is confirmed that colony patterns in the region D are consistent with the behavior of solutions of (40) [16]. As a natural extension of (40), there is a model for the nutrient-limited growth of bacteria. It is given by bt = db ∆b + ωg(n)b, nt = dn ∆n − g(n)b,

(41)

where n is the concentration of nutrients, dn is the diffusion coefficient of the nutrients, g(n) is the growth rate. A simple form is the Malthusian rate g(n) = n. The positive constant ω is the conversion rate of nutrients from intake to growth. The initial conditions for (41) are b(0, x) = b0 (x), n(0, x) = n0 ,

(42)

where b0 (x) is a delta-like distribution of the initial density of the bacteria and n0 is the initial concentration of nutrients which is distributed uniformly in space. We note that the parameters (n0 , db ) in the model (41), (42) correspond to (Cn , Ca−1 ) in the experiments. However, even if two parameters are globally varied, the resulting patterns generate disk-like patterns only. In order to improve this model so as to generate branched patterns, Kawasaki et al. [17] proposed a nonlinear diffusion model which is described by bt = ∆(σ(x, y)nb∆b) + nb, nt = dn ∆n − nb,

(43)

where b and n are the same as the ones in (41) and σ(x, y) is a diffusion coefficient with small spatial fluctuations. Furthermore, along the same line as (41), Kitsunezaki [18] proposed a density-dependent diffusion model including the death term

104

Masayasu Mimura

bt = ∆(db b∆b) + nb − αb, nt = dn ∆n − nb,

(44)

where db and α are positive constants. Bacterial corpse, say w(t, x), is given by wt = αb. Numerical simulations of these two models reproduce branched patterns ranging from DBM-like patterns to disk-like patterns. From the modelling perspective, we still have the following problem Is there any unified model which generates all of growth patterns observed in experiments when two parameters corresponding to Cn and Ca are varied globally? For this problem, we propose a continuous model for the density of bacterial cells and the concentration of nutrient at the macroscopic level [19]. The essential ansatz in our modelling is the introduction of the internal state in each cell, so that, if this state variable is increasing, then the bacteria actively move, grow, and perform cell-division; while if it is decreasing, then it does nothing at all. In order to model this situation in a simple way, we assume that the bacterial cells consist of two types: active cells and inactive ones. In fact, a group of actively moving cells is clearly observed in the tip of each growing finger in the region E, as if they were a finger nail (Fig. 9).

Fig. 9. Caricature of finger-nail pattern

They seem to drive the growth of the finger tip, leaving inactive cells behind. Let b(t, x) and s(t, x) be the concentrations of the active and inactive cells, respectively, (hence the sum b(t, x) + s(t, x) is the concentration of all bacterial cells) and n(t, x) is the concentration of nutrients. The model for b, s and n is given by bt = ∆(db ∆b) + ωg(n)b − a(b, n)b, nt = dn ∆n − g(n)b, st = a(b, n)b,

(45)

where db = d(k) and dn are the diffusion rates of the bacterial cells and the nutrient, respectively. Here k is the parameter corresponding to agar concentrations. A plausible assumption is that d(k) is monotone decreasing with respect to k. A simple form of d(k) is d(k) = d0 /(1 + k), with some positive constant d0 . The growth rate of active cells is ωg(n). We assume g(n) to be

Reaction-Diffusion Systems

105

an exponential (or Malthusian) growth rate of cells. The rate of change from the active cells to the inactive ones is a(b, n), which is specified by the mechanism of the internal state. We neglect the rate of change from the inactive cells to the active ones, because it is observed that once active cells become inactive, they never become active again unless food is added artificially. It is plausible that a( · , n) decreases with an increase in nutrient concentration n. However, the dependency of a(b, · ) on the active cell population b is still unclear. Since it is observed that, if the cell population b becomes quite small, each cell is not very active, we assume that a(b, · ) is a decreasing function of b. Hereafter, for concreteness and simplicity, we adopt the functional form a(b, n) = a0 (1 + a1 b)−1 (1 + a2 n)−1 with positive coefficients ai (i = 0, 1, 2). It should be stressed, however, that this particular form is not essential to our conclusion below, and other functions which decrease with b and n will work as well. We rewrite the first two equations of (45) as the following nondimensionalized form bt = d∆b + nb − a(b, n)b, nt = ∆n − nb,

(46)

where d is the ratio of the diffusion rates db and dn . It is noted that these two equations (46) are closed for b and n. The initial conditions for (46) are b(0, x) = b0 (x), n(0, x) = n0 ,

(47)

which turn out to be the same as (42). Consequently, we can find that the RD model (46) is a slight modification of (41). The only difference is the presence of the conversion term a(b, n)b in the first equation. The third variable s can be obtained in terms of b and n by solving st = a(b(t, x), n(t, x))b(t, x), s(0, x) = 0.

(48) (49)

We start with showing the following theorem (cf. [20]) Theorem 4.1. Consider (46), (47) in a bounded domain Ω under zero flux boundary conditions. Then there is some constant n∞ > 0 such that lim (b(t, x), n(t, x)) = (0, n∞ ).

t→∞

This theorem says that the active cells density b(t, x) does not generate any pattern, even simple disk-like patterns never occur. This conclusion seems to give negative information to us. However, what we want to know is how the total bacterial density b(t, x) + s(t, x) forms spatial patterns. The above theorem also tells that there is s∞ (x) ≥ 0 such that lim (s(t, x)) = s∞ (x).

t→∞

106

Masayasu Mimura

Therefore, we find lim (b(t, x) + s(t, x)) = s∞ (x),

t→∞

which gives colony patterns we want to know. Unfortunately, there is no analytic methods to do it at present, so that we rely on numerical methods. Let us solve (46)-(49) in a two-dimensional rectangular domain, taking n0 and d as free parameters. For quite small n0 and d (hard and poor environment), one point distribution of b(x) expands and breaks up into several spots, and then each spot splits into two smaller ones repeatedly as time goes on so as to generate many smaller spots which appear but eventually fade out. It is noted that splitting process of a single spot of b corresponds to tip-splitting process of branched patterns of s. Through this process, b(t, x) + s(t, x) exhibits very complex branched patterns, which resemble colony patterns observed in the region A (Fig. 10).

Fig. 10. Numerical branched pattern in (46)-(49)

The fractal dimension is about 1.67. These ehibit characteristic of DLA growth. Increasing d with the same value of n0 , the pattern changes to a DBM-like pattern. On the other hand, for large d and n0 (soft and rich environment), b(t, x) shows an expanding ring pattern where no unstable mechanism occurs, so that the pattern of b(t, x) + s(t, x) simply forms an expanding disk which is similar to the ones in region D. If d is decreasing with fixed large n0 , a pattern of growth and no growth occurs alternately and the ever-expanding disk pattern changes to an immobilized concentric-ring pattern. In spite of the simplicity of the system, it is surprising that the model (46)-(49) reproduces four qualitatively different patterns corresponding to those in the regions A, C, D, and E, depending on values of d and n0 . Thus, summarizing the above results, our reaction-diffusion model exhibits the phase diagram of colony patterns, as in Fig. 11.

Reaction-Diffusion Systems

107

Fig. 11. Phase diagram of numerical colony patterns in (46)-(49)

For Eden-like patterns in region B (hard agar medium with rich nutrient), experimental observations indicate that bacterial cells hardly move, but cells grow moderately because of an adequate supply of nutrient, so that the mechanism of expansion of colony is different from the others. We will not touch with this (see [20]). 4.2 Grey-Scott model Theoretical understanding of spatio-temporal patterns arising in far from equilibrium states have been investigated by using RD systems describing autocatalytic chemical reactions with feed processes. For example, there is the following cubic autocatalytic chemical reaction processes 2V + U → 3V, V → W. Letting u, v and w be the concentrations of the chemical species U, V and W , the model describing the above processes is ut = du ∆u − uv 2 + F (1 − u), vt = dv ∆v + uv 2 − (F + k)v,

(50)

where du and dv are diffusion coefficients of U and V , respectively, and parameters F ≥ 0 and k > 0 are constants. (50) is called the Grey-Scott model [21]. When F > 0, (50) provides feeding processes so that it is called an open system of the Grey-Scott model, while, when F = 0, (50) is ut = du ∆uv 2 − uv 2 , vt = dv ∆v + uv 2 − kv,

(51)

which is called a closed system. For (51) in a bounded domain with zero-flux boundary conditions, it is known that there is some constant u∞ > 0 such that

108

Masayasu Mimura

lim (u(t, x), v(t, x)) = (u∞ , 0).

t→∞

This result indicates that any solution (u, v) tends to be spatially homogeneous for large time, that is, there occurs no pattern formation. This result can be intuitively understood, because the system does not provide feeding process. On the other hand, for F > 0, it is numerically demonstrated that (50) generates a surprising variety of spatio-temporal patterns, sensitively depending on parameters F and k [22]. For these results, we had long believed that the reason why such variety of patterns occur is that (50) is an open system, in other words, it is far from equilibrium system, and the closed system (51) is less interesting from pattern formation. However, we should notice that (n, b) in the bacterial colony model (46) possesses similar interaction to (u, v) in (51). Assume that the product W is immobile so that the resulting equation for w is wt = kv.

(52)

This indicates that w corresponds to s in (48), which suggests that w exhibits complex branched patterns. In order to confirm it, we write (51) in twodimension rectangular domain Ω with zero flux boundary conditions ∂u/∂ν = ∂v/∂ν,

t > 0, x ∈ ∂Ω,

(53)

where ν is the outward normal unit vector on the boundary ∂Ω. The initial conditions are u(0, x) = u0 (x) ≥ 0,

v(0, x) = v0 (x) ≥ 0,

w(0, x) = 0 x ∈ Ω.

(54)

Here we simply assume that u0 (x) = u0 is constant in Ω. In a way similar to that of the theorem for the bacterial colony model, we can prove for (51)-(54) that there is some function w∞ (x) ≥ 0, depending on both u0 and v0 , such that lim w(t, x) = w∞ (x). (55) t→∞

Spatial pattern of w∞ (x) is numerically studied, by taking d and u0 as free parameters. For small d and u0 , as is expected, v(t, x) generates several spots and each small spot irregularly moves outward, splitting into two smaller spots. Of course, after large time, all the spots fade out (Fig. 12-(a)). On the other hand, w(t, x), which is given by
w(t, x) = k v(τ, x)dτ, x ∈ Ω, t

records the history of v and exhibits a complex branched pattern, as shown in Fig. 12-(b). One can thus understand that the dynamics of v and w quitely resembles the ones of n and b in the bacteria colony model (46).

Reaction-Diffusion Systems

Fig. 12. (a) Spot-splitting of v(t, x)

109

(b) Tip-splitting of w(t, x)

Generally speaking, these systems fall in the framework of limited resource-comsumer RD systems. Quite recently, it has been reported that this kind of complex pattern also appears in experiments on combustion in micro-gravity [23]. Furthermore, we find that this models can be described by a limited resource-comsumer RD system. We now conclude that, though any solution of a class of RD systems with limited resource-consumer interaction tends to be spatially homogeneous, the history of the consumer such as active bacteria s in (48) and the final product w in (52) forms very complex branched patterns. It turns out that the study of this class of systems is also an interesting problem from pattern formation viewpoint, even if there are closed systems.

5 Competition-diffusion systems and singular limit procedures Understanding of spatial and/or temporal behaviors of ecologically interacting species is a central problem in population ecology. As for competitive interaction of ecological species, problems of coexistence or exclusion have been theoretically investigated by using different types of mathematical models [24]. Especially, a variety of RD equations have been proposed to study spatial segregation of competing species [25]. Quite recently, in order to understand the evolutional behavior of spatially segregating regions of competing species, the methods called spatial segregation limits have been successfully developed. These enable us to derive evolutional equations describing the boundaries of spatially segregating regions of competing species. In some situations, the derived equations are described by new types of free boundary problems. As one of the well-known models, we consider a RD system of GauseLotka-Volterra type. Let ui (t, x) be the population density of the i-th species

110

Masayasu Mimura

Ui (i = 1, 2, . . . , N ) at time t > 0 and the position x ∈ Ω, where Ω is a bounded domain in R2 . The resulting system for ui (i = 1, 2, . . . , N ) is given by t > 0, x ∈ Ω. (56) " Here di is the diffusion coefficient of ui and fi = (ri − i,j aij uj )ui , where ri is the intrinsic growth rate, aii and aij are respectively the intraspecific and the interspecfic competition rates (i, j = 1, 2, . . . , N ) . All of the rates are positive constants. We impose zero-flux boundary conditions to (56) uit = di ∆ui + fi (u1 , u2 , . . . , un )

(i = 1, 2, . . . , N ),

∂ui /∂ν = 0 (i = 1, 2, . . . , N ),

t > 0,

x ∈ ∂Ω,

(57)

where ν is the outerward normal unit vector on ∂Ω. The initial conditions are ui (0, x) = u0 i(x)

(i = 1, 2, . . . , N ),

x ∈ Ω.

(58)

The simplest system of (56) is the case with N = 2, that is, u1t = d1 ∆u1 + (r1 − a1 u1 − b1 u2 )u1 u2t = d2 ∆u2 + (r2 − b2 u1 − a2 u2 )u2

,

t > 0, x ∈ Ω.

(59)

With the same boundary and initial conditions as (57), (58), qualitative properties of non-negative solutions of (59) have been intensively studied in mathematical cummunities. The first remark is that the stable attractor of (59) consists of equilibrium solutions only [1]. By this information, one finds that existence and stability of non-negative equilibrium solutions play an important role on the study of asymptotic behavior of solutions. We ecologically assume the situation where two species are strongly competing, that is, a1 /b2 < r1 /r2 < b1 /a2 , (60) for which, stable spatially constant equilibrium solutions are (u1 , u2 ) = (r1 /a1 , 0) and (0, r2 /a2 ) only. These solutions mean that one of the competing species survives and the other is extinct. When the domain Ω is convex, all spatially non-constant equilibrium solution is unstable, even if they exist [26]. In other words, stable equilibria are only (r1 /a1 , 0) and (0, r2 /a2 ), which ecologically indicates that two strongly competing species can never coexist in convex habitats. That is, if the domain is convex, the asymptotic behavior is extremely simple. However, if Ω is not convex, the structure of equilibrium solutions is not so simple but depends on the shape of Ω. If Ω takes suitable dumb-bell shape, for instance, there exist stable non-constant equilibrium solutions, which exhibit spatial segregation of the two competing species in a sense that one region is nearly occupied by the species U1 , while the other does by U2 , that is, two competing species possibly coexist, if the habitat is suitably nonconvex [27].

Reaction-Diffusion Systems

111

For multi-competing species case, the situation is drastically different from the case N = 2. Even if we restrict ourselves to the diffusionless system of (56), the solution-structures are rather complicated. There is possible coexistence of a variety of periodically and aperiodically solutions, in addition to equilibria, depending on parameters ri and aij (for instance, [28]). There has been little work for the RD system (56) with N ≥ 3, except for some results which are discussed by using singular perturbation methods [29]. 5.1 Spatial segregating limits of two competing species model The asymptotic behavior of solution (u1 (t, x), u2 (t, x)) to (59) is already known. If the domain is convex, the solution tends to either (r1 /a1 , 0) and (0, r2 /a2 ). Even if we know this result, an ecologically important problem so as to know the transient behavior of solutions is still unsolved. For this purpose, the methods which are called spatial segregation limits have been developed. Assume that either the diffusion coefficients d1 and d2 are sufficiently small or all of the other rates ri , ai , and bi (i = 1, 2) are sufficiently large. Then (59) is rewritten as u1t = ε2 ∆u1 + (r1 − a1 u1 − b1 u2 )u1

,

u2t = dε2 ∆u2 + (r2 − b2 u1 − a2 u2 )u2

t > 0, x ∈ Ω,

(61)

in which ε is a small parameter. Under the assumption (60), it is expected that there occur transition sharp layer regions which separate two subregions Ω1 (t) = {x ∈ Ω | (u1 (t, x), u2 (t, x)) ≈ (r1 /a1 , 0)} and Ω2 (t) = {x ∈ Ω | (u1 (t, x), u2 (t, x)) ≈ (0, r2 /a2 )}, which ecologically mean segregating regions of competing species. To study the time evolution of Ωi (t) (i = 1, 2), we take the limit as ε ↓ 0 in (61), so that the layers become interfaces, say Γ (t), which represent the boundary between two different regions Ω1 (t) and Ω2 (t). Using the singular limit analysis, the following evolution equation to describe the motion of the interface Γ (t) can be derived as follows [30] ∇(t) = c − εL(d)κ,

(62)

where V = ε∇ is the normal velocity of the interface. L(d) is a positive constant depending on d such that L(1) = 1 and c is the velocity of the one dimensional travelling wave solution (u1 , u2 ) of u1t = u1xx + (r1 − a1 u1 − b1 u2 )u1 u2t = du2xx + (r2 − b2 u1 − a2 u2 )u2

,

t>0

− ∞ < x < ∞,

(63)

with the boundary conditions (u1 , u2 )(t, −∞) = (r1 /a1 , 0) and (u1 , u2 )(t, ∞) = (0, r2 /a2 ).

(64)

For (63), (64), the velocity c of the travelling wave solution (u1 (z), u2 (z)) (z = x − ct) is unique for fixed values of ri , ai and bi (i = 1, 2) [31]. It is very

112

Masayasu Mimura

interesting that the interface equation (62) is the same as (31) for the scalar bistable RD equation (30). This result shows that qualitative properties of solutions of (62) and the scalar bistable RD equation (30) are quite similar in the limit ε ↓ 0. There is another spatial segregating limit. It is the situation when the interspecific competition rates b1 and b2 are very large [32]. To study this situation, it is convenient to rewrite (59) as u1t = d1 ∆u1 + r1 (1 − u1 )u1 − bu1 u2 u2t = d2 ∆u2 + r2 (1 − u2 )u2 − αbu1 u2

,

t > 0, x ∈ Ω,

(65)

where b and α are positive constants. We assume that b is the only parameter which is large and that all other parameters are of order O(1). The coefficient α is the competition rate between two species u1 and u2 . If α > 1, then u1 has a competitive advantage over u2 , while if α < 1, the situation is reversed. When b is sufficiently large, in other words, the interspecific competiton is very strong, one can expect that u1 and u2 have disjoint supports (habitats) with only curves which separate the habitats of the competing species. We now derive the time evolution equation of the supports of u1 and u2 , taking the limit as b ↑ ∞. The limiting system can be described by a free boundary problem which is a two phase Stefan-like problem with reaction terms [33]. Let Γ (t) be the interface which separates the two subregions Ω1 (t) = {x ∈ Ω | u1 (t, x) > 0, u2 (t, x) = 0} and Ω2 (t) = {x ∈ Ω | u1 (t, x) = 0, u2 (t, x) > 0} in Ω. Then u1 and u2 satisfy equations u1t = d1 ∆u1 + r1 (1 − u1 )u1 , u2t = d2 ∆u2 + r2 (1 − u2 )u2 ,

t > 0, x ∈ Ω1 (t), t > 0, x ∈ Ω2 (t)

(66)

and boundary conditions ∂ui /∂ν = 0 (i = 1, 2),

t > 0, x ∈ ∂Ω.

(67)

On the interface Γ (t), we have u1 (t, x) = 0,

u2 (t, x) = 0,

t > 0, x ∈ Γ (t)

(68)

t > 0, x ∈ Γ (t),

(69)

and 0 = −αd1 ∂u1 /∂ζ − d2 ∂u2 /∂ζ ,

where ζ is a unit vector normal to Γ (t). The initial conditions are given by ui (0, x) = ui0 (x)

(i = 1, 2),

x ∈ Ω,

(70)

and are such that their support is separated by Γ (0) = Γ0 .

(71)

The problem is to find (u1 (t, x), u2 (t, x)) and Γ (t) which satisfy (66)-(71). If this problem can be solved, the interface Γ (t) determines segregating patterns between the two strongly competing species.

Reaction-Diffusion Systems

113

5.2 Dynamics of triple junctions arising in three competing species model We consider (22) with N = 3 in two dimensions, that is, uit = di ∆ui + fi (u1 , u2 , u3 ),

t > 0, x ∈ Ω ⊂ R2 ,

(72)

where fi (u1 , u2 , u3 ) = (ri − ai1 u1 − ai2 u2 − ai3 u3 )ui (i = 1, 2, 3), or simply t > 0, x ∈ Ω,

Ut = D∆U + F (U ),

(73)

where U = (u1 , u2 , u3 ), D is the 3 × 3 diagonal matrix with positive elements {di } (i = 1, 2, 3), and F = (f1 , f2 , f3 ). We first consider the diffusionless system corresponding to (73) Ut = F (U ),

t > 0.

(74)

Assume that aij (i = j) are large comparing with others aii , in order to require that P1 = (r1 /a11 , 0, 0), P2 = (0, r2 /a22 , 0) and P3 = (0, 0, r3 /a33 ) are stable and other critical points are all unstable, that is, three species are in strong competition. It is thus shown that almost all non-negative solutions U (t) of (74) tend to one of Pi (i = 1, 2, 3) [34]. This ecologically implies that one of the competing species can survive and the other two are extinct. This result indicates that the dynamics of U (t) of (74) is extremely simple. Under the non-coexistence situation stated above, we consider what kind of patterns RD system (74) generates by the interplay of diffusion and reaction. In order to study this problem, we assume that the diffusion rates D are sufficiently ¯ with a sufficiently small parameter ε, where D ¯ is a small, that is, D = εD diagonal matrix with elements {d¯i } (i = 1, 2, 3). The resulting system from (73) is written as Ut = ε2 D∆U + F (U ),

t > 0, x ∈ Ω,

(75)

¯ as the original D. The initial and boundary conditions are where we wrote D U (0, x) = U0 (x),

x∈Ω

(76)

and ∂U/∂ν = 0,

t > 0, x ∈ ∂Ω,

(77)

where ν is the outward normal unit vector on ∂Ω. Since ε is sufficiently small, one can expect that the behavior of solutions of (75)-(77) essentially consists of two stages. The first stage is the occurrence of transition layer regions (when ε tends to zero, they become interfacial curves), which generally divide the domain Ω into three subdomains Ω1 , Ω2 , and Ω3 , where the solution U (t, x) is close to one of Pi (i = 1, 2, 3). This indicates the appearance of spatial segregation of three competing species with triple junctions, where

114

Masayasu Mimura

three interfacial curves meet. The second stage is the motion of interfacial curves with triple junctions, that is, the dynamics of segregating patterns. We numerically consider (75)-(77) in a rectangular domain Ω. As was stated above, the first stage is the occurrence of segregating patterns of (u1 , u2 , u3 ) so that the domain Ω is clearly divided into three subdomains Ω1 , Ω2 and Ω3 , which are separated by interfacial curves with triple junctions (Fig. 13).

Fig. 13. Occurrence of transition layers among three segregating regions

Next consider the second stage. The first we will discuss is the completely symmetric case where di = d, aii = a, aij = b (i = 1, 2, 3). The dynamics of segregating pattern changes slowly (Fig. 14).

Fig. 14. Segregation regions with triple junctions

Here we note two points (i) if interfacial curves are almost straight, they move very slowly; (ii) angles between any two neighboring interfacial curves are almost equal. This angle condition of interfaces can be intuitively understood by the fact that three competing species possess completely symmetric property. The next cases are non-symmetric. The first is a semi-symmetric case where di = d (i = 1, 2, 3), a12 = a21 = b1 , a23 = a32 = b2 , a31 = a13 = b3 , but b1 , b2 and b3 are not necessarily equal. The dynamics of segregating pattern is qualitatively similar to the above symmetric case except for the angles of interfaces at triple junction areas, which seem to be not necessarily equal. The second is the case with ordering property where a12 < a21 , a13 < a31 and a23 < a32 , that is, the species U1 is the strongest of the three. The resulting dynamics of pattern is much faster than the previous two cases

Reaction-Diffusion Systems

115

and, as is easily expected, the domain is gradually occupied by the strongest species U1 . The third is a symmetrically cyclic case where a12 = a23 = a31 = b and a21 = a32 = a13 = b with b ≈ b . As the cyclic property suggests, one can expect that there appear steadly rotating spiral patterns with three arms (Fig. 15).

Fig. 15. Perfect spiral waves with three arms

Finally, the fourth is a general case with cyclic property where a12 < a21 , a23 < a32 and a31 < a13 . There is no longer any stationary rotating spiral but complex spatio-temporal pattern with several clustering spirals, where each spiral seems to be steadily rotating in a vicinity of triple junction areas (Fig. 16). The precise discussion is stated in [35].

Fig. 16. Clusters of spiral waves

We have observed that segregating patterns of three competing species are drastically different from the ones of two competing species, that is, even if Ω is convex domain in R2 , it is possible for the three species to coexist but the resulting pattern is not stationary but very dynamic.

6 Chemotactic patterns - aggregation and colonies In the movement of biological individuals, it is often observed that they have a tendency to aggregate by moving preferentially toward higher concentrations of chemotactic substances. As an example, Dictyostelium discoideum (Dd) is a kind of microorganisms which show aggregation of single cells to form cellular aggregates, as if it were amoebae, at some phase in their life cycles. It

116

Masayasu Mimura

is observed that multicellular development is induced by lack of nutrients and leads to the formation of a motile slug-like organism, eventually transforming into a fruiting body. Following starvation, they acquire the ability to respond to extracellular substance, which is called cAMP with intracellular synthesis and secretion of cAMP and also with chemotactic movement towards higher gradient of cAMP concentration. RD model which takes chemotactic effect into account was originally proposed by Keller and Segel [36]. It is described by nt = dn ∆n − ∇(n∇χ (u)), ut = du ∆u + h(u, n),

(78)

where n(t, x) is the population density of individuals and u(t, x) is the concentration of chemotactic substance at time t and position x. dn and du are respectively the diffusion coefficients of n and u. χ(u) is a sensitivity function of chemotaxis and ∇χ(u) is the velocity of the direct movement of n due to chemotaxis, where χ(u) generally satisfies χ (u) ≥ 0 for u > 0. Some plausible forms of χ(u) are proposed such as χ(u) = q log u,

χ(u) = qu, and χ(u) = qu2 (s2 + u2 )−1 ,

with positive constants q and s. h(u, n) is the reaction term leading to production and degradation of u so that ∂h/∂n > 0 and ∂h/∂u < 0 are assumed for any u > 0 and n > 0. A simplest form of h is h(u, n) = βn − γu with positive coefficients β and γ. The system (78) implies that individuals move by diffusion and chemotaxis and the total number of individuals is preserved, because neither growth nor death occur in the process. Under the concept of chemotaxis-induced instability, there exist stable non-constant equilibrium solutions, which indicate chemotactic aggregation of individuals, in a bounded domain with zero flux boundary conditions [37]. Furthermore, the blow-up problem, which indicates chemotactic collapse, is discussed in higher dimensions [38]. 6.1 Aggregation of Dictyostelium discoideum Following starvation, Dd acquire the ability to respond to extracellular substance, which is called cAMP with intracellular synthesis and secretion of cAMP and also with chemotactic movement towards higher gradient of cAMP concentration. Concentric and spiral waves of cAMP are observed and cell movement towards the centers of the wave patterns takes place. However, cell movement subsequently ceases and the wave patterns break up and amoebae organize in a pattern of branching cell stream. Several continuous models describing aggregation process of Dd have been proposed so far. They are basically described by the following three component system ([39], [40], for instance)

Reaction-Diffusion Systems

ut = d∆u + λa(n)f (u, v), vt = g(u, v), nt = D∆n − ∇(χ(v)n∇u),

117

(79)

where u, v and n denote the concentration of extracellular cAMP, the fraction of active cAMP cell membrane receptors per cell and the cell density (number of cells per area), respectively. f (u, v) is the rate of cAMP synthesis and degradation per cell. The total production per unit area depends on the cell density with a(n). g(u, v) is the kinetics of desensitization and the recovery of the active form is the ratio of the characteristic rates for cAMP synthesis and receptor desensitization. When the cell density is constant in space and time, that is, n is constant, the first two equations for u and v are qualitatively similar to activator-inhibitor typed RD systems (20), which was introduced by Turing. The third equation for n describes the dynamics of cell movement with diffusivity D and the chemotactic movement with the velocity χ(v)∇u, which is similar to (78). Numerical computation of (79) demostrates that pattern formation consists of two stages: the first stage is the occurrence of spiral waves. Its mechanism is similar to the one suggested by Hodgkin and Huxley and the second is broken up of spiral waves into streaming patterns, which is chemotaxis-induced instability. We thus find that the ideas by Turing, Hodgkin and Huxley are included in the model (79). 6.2 New pattern arising in a chemotaxis-diffusion-growth system In this section, we consider the following two component system including diffusion, chemotaxis and growth nt = dn ∆n − ∇(n∇χ(u)) + f (n), ut = du ∆u + βn − γu,

(80)

where f (n) is the growth term of the form f (n) = (g(n)−α)n with the growth rate g(n) including cooperation and competition effects and with the degradation rate α due to exterior forces such as predation and/or intoxication. By suitable transformations, (80) can be written as nt = dn ∆n − ∇(n∇χ(u)) + f (n), ut = du ∆u + n − γu.

(81)

We assume that f (n) = n(1 − n)(n − a) with small a > 0. We first that (81) has three constant equilibrium solutions (n, u) = (0, 0), (1, 1/γ) and (a, a/γ); the first two are stable, while the last is unstable, that is, (81) is a bistable system. The boundary condition is lim (n(t, x), u(t.x)) = (0, 0),

|x|→∞

t > 0.

(82)

118

Masayasu Mimura

In the absence of chemotaxis, (81) reduces to (24) for which, if n(0, x) is a relatively large delta-like function, it expands uniformly and forms a disklike pattern. We now address the following question: if the chemotactic effect becomes stronger, are there any other patterns generated by (81)? In order to answer to this question, we assume the situation where (i) the movement of individuals is mainly due to chemotaxis, (ii) the chemotactic substance diffuses so fast compared with the migration of individuals. We thus introduce a small parameter ε into (81) and rewrite it as nt = εdn ∆n − k∇(n∇χ(u)) + ε−1 f (n) εut = du ∆u + n − γu

,

t > 0, x ∈ R2 ,

(83)

where k is a positive constant such that χ(u) is normalized to satisfy maxu>0 χ (u) = 1. It is first noted that, when k = 0, the first of (83) reduces to (30) so that there appear transition layers between two states where u takes values nearly 1 and 0, and furthermore, in the limit ε ↓ 0, the aggregating region is described by the resulting interface Γ (t) and the time evolution of Γ (t) is V (t) = c(a) − εκ, t > 0, where V (t) is the normal velocity of the interface Γ (t) and κ is the curvature of the interfacial curve [41]. As a situation similar to (30), system (83) generates transional layers between two states: u takes values nearly 1 and 0. In ecological terms, the region where u is near 1 results an aggregating region of individuals. Moreover, in the limit as ε ↓ 0, the resulting interface Γ (t) is described by V (t) = c(a) − εκ − kχ (uI ), t > 0, where uI is the value of u on the interface Γ (t) [41]. Noting that c(a) > 0, because a is small, and χ (u) > 0, one can expect the appearance of localized solutions under suitable balance between the growth and chemotaxis. In fact, singular perturbation analysis shows that there exist radially symmetric equilibrium solutions in a suitable range of k for sufficiently small but not zero ε. Moreover, the stability of this solution can be discussed. If χ (u) ≤ 0, it is stable, while if χ (u) > 0, the stability depends the values of d (for a precise discussion, see [40]). With the sensitive function χ(u) = qu2 (s2 + u2 )−1 w, we only show numerical simulations of solutions which are destabilized from radially symmetric equilibrium solutions, as shown in Fig. 17. Typical patterns are a sponge pattern (Fig. 17-(a)), network one (Fig. 17-(b)) and fingering one (Fig. 17-(c)), depending on values of k.

Reaction-Diffusion Systems

Fig. 17. (a) Sponge pattern

(b) Network pattern

119

(c) Fingering pattern

References [1]

[2]

[3] [4] [5]

[6] [7] [8] [9]

Hirsch, M.W., Differential equations and convergence almost everywhere of strongly monotone flows, Contemp. Math., 17, Amer. Math. Soc., 267-285 (1983) Casten, R.G. and Holland, C.J., Instability results for reaction-diffusion equations with Neumann boundary conditions, J. Differential Equations, 27, 266273 (1978) Matano, H., Asymptotic behavior and stability of solutions of semilinear parabolic equations, Pub. R.I.M.S., Kyoto Univ., 15, 401-454 (1979) Turing, A., The chemical basis of morphogenesis, Phil. Trans. Roy. Soc. Lond. B, 237, 23-72 (1952) Hodgkin, A.L. and Huxley, A.F., A quantitative description of membrane and its application to conduction and excitation in nerve, J. Physiol., 117, 500-544 (1952) de Mottoni, P. and Schatzman, M., Development of interfaces in RN , Proc. Royal Soc. Edinburgh Sect. A, 116, 207-220 (1990) Grayson, M., The heat equation shrinking convex plane curves, J. Differential Geometry, 26, 285-314 (1987) Carr, J. and Pego, R.L., Metastable patterns in solutions of ut = ε2 uxx −f (u), Comm. Pure Appl. Math., 42, 523-576 (1989) Evans, L.C., and Spruck, J. , Motion by level sets by mean curvature I, J. Differential Geometry, 33, 635-681 (1991)

120

Masayasu Mimura

[10] Chen, Y.G., Giga, Y. and Goto, S., Uniqueness and existence of viscosity solutions of generalized mean curvature flow equations, J. Differential Geometry, 33, 749-786 (1991) [11] Ikeda, H., Mimura M. and Nishiura, Y., Global bifurcation phenomena of travelling wave solutions for some bistable reaction-diffusion systems, Nonlinear Anal. Theory, Methods & Appl., 13, 507-526 (1989) [12] Nishiura, Y., Ikeda, H., Mimura, M. and Fujii, H., Singular limit analysis of stability of traveling wave solutions in bistable reaction-diffusion systems, SIAM J. Math. Anal., 21, 85-122 (1990) [13] Matsushita, M. and Fujikawa, H., Diffusion-limited growth in bacterial colony formation, Physica A, 168, 498-506 (1990) [14] Vicsek, T., Fractal growth phenomena, 2nd edition. World Scientific, Singapore (1992) [15] Ben-Jacob, E. et al., Generic modelling of cooperative growth patterns in bacterial colonies, Nature, 368, 46-49 (1994). [16] Wakita, J. et al., Experimental investigation on the validity of population dynamics approach to bacterial colonial formation, J. Phys. Soc. Japan, 63, 1205-1211 (1994). [17] Kawasaki K., et al., Modelling spatio-temporal patterns generated by Bacillus subtilis, J. Theoret. Biol., 188, 175-185 (1997) [18] Kitsunezaki S., Interface dynamics for bacterial colony formation, J. Phys. Soc. Japan, 66, 1544-1550 (1997) [19] Mimura, M., Sakaguchi, H. and Matsushita, M., A mathematical model of bacterial colony patterns, Proceedings of Kyoto Conference on Mathematical Biology (1996) [20] Mimura, M., Sakaguchi, H. and Matsushita, M., Reaction-diffusion modelling of bacterial coloniy patterns, Physica A, 282, 283-303 (2000) [21] Gray, P. and Scott, S.K., Autocatalytic reactions in the isothermal, continuous stirred tank reactor: oscillations and itstabilities in the system A + B → 3B, B → C, Chem. Eng. Sci., 39, 1087-1097 (1984) [22] Pearson, J.E., Complex patterns in a simple system, Science, 261, 189-192 (1993) [23] Lubkin, G.B., Combustion in two dimensions yields fingering instability, Physics Today, Jan., 19-21 (1999) [24] Gause, G.F., The Struggle for Existence, The Williams & Wilkins Company, Baltimore, 1934. [25] Okubo A. et al., On the spatial spread of the grey squirrel in Britain, Proceedings of the Royal Society of London, Series B, 238, 113-125 (1989) [26] Kishimoto, K. and Weiberger, H., The spatial homogeneity of stable equilibria of some reaction-diffusion systems on convex domains, J. Differential Equations, 58, 15-21 (1985) [27] Matano, H. and Mimura, M., Pattern formation in competition-diffusion systems in non-convex domains, Pub. R.I.M.S., Kyoto Univ., 19, 1049-1079 (1983) [28] May. R and Leonard, W.J., Nonlinear aspects of competition between species, SIAM J. Appl. Math., 29, 243-275 (1975) [29] Mimura, M. and Fife, P., A 3-component system of competition and diffusion, Hiroshima Math. J., 16, 189-207 (1986) [30] Ei, S.-I. and Yanagida, E., Dynamics of interfaces in competition-diffusion systems, SIAM J. Appl. Math., 54, 1355-1373 (1994)

Reaction-Diffusion Systems

121

[31] Kan-on, Y., Parameter dependence of propagation speed of travelling waves for competition-diffusion equations, SIAM J. Math. Anal., 26, 340-363 (1995) [32] Dancer, E.N., Hilhorst, D., Mimura, M. and Peletier, L.A., Spatial segregation limit of a competition-diffusion system, European J. Appl. Math., 10, 97-115 (1999) [33] Ockendon, J.R. and Hodgkins, W.R., eds., Moving boundary problems in heat flow and diffusion, Oxford Univ. Press, Oxford (1975) [34] Van den Driessche, P. and Zeeman, M.L., Three-dimensional competitive Lotka-Volterra systems with no periodic orbits, SIAM J. Appl. Math., 58, 227-234 (1998) [35] Ei, S.-I., Ikota, R. and Mimura, M., Segregating partition problem in competition-diffusion systems, Interface and Free Boundaries, 1, 57-80 (1999) [36] Keller, E.F. and Segel, L.A., Initiation of slime mold aggregation viewed as an instability, J. Theor. Biol., 26, 399-415 (1970) [37] Schaaf, R., Stationary solutions of chemotaxis systems, Trans. Amer. Math. Soc., 292, 531-556 (1985) [38] Herrero, M.A. and Velazquez, J.J.L., Chemotactic collapse for the Keller-Segel model, J. Math. Biol., 35, 177-194 (1996) [39] Hoefer, T., Sherratt, J.A. and Maini, P.K. Cellular pattern formation during Dictyosteleium aggregation, Physica D, 85, 425-444 (1995) [40] Vasiev, B. NB., Hogeweg, P.H. and Panfilov, A.V., Simulation of Dictyosteleium Discideum aggregation vis reaction-diffusion model, Phys. Rev. Letters, 73, 3173-3176 (1994) [41] Mimura, M. and Tsujikawa, T., Aggregating pattern dynamics in a chemotaxis model including growth, Physica A, 230, 499-543 (1996)

Lectures on Evolution Free Boundary Problems: Classical Solutions Vsevolod A. Solonnikov Saint-Petersburg Department of Steklov Mathematical Institute, Fontanka, 27, 191011 S-Petersburg, Russia [email protected]

Introduction The aim of these notes is to present some methods of the proof of the solvability of free boundary problems for the parabolic and nonstationary Navier– Stokes equations. As “test problems” we have chosen the one-phase Stefan problem and the problem of evolution of an isolated liquid mass. These methods allow us to show that the above problems (and some others to which they are applicable) possess unique classical solutions defined on some finite time intervals. In addition, we prove that the second problem is solvable in an infinite time interval t > 0, when initial data are close to an equilibrium state. Of course, for one-phase Stefan problem, this result is far from being optimal, since the free boundary may be analytical for analytical data (see [18] and more references in the survey [28]). Additional references and results on the Stefan problem, including weak solutions, may be found, for instance, in the books [23], [15] or [45]. The proof consists in the transformation of the free boundary problems into nonlinear evolution problems in given domains and in a careful analysis of linearized problems. Sharp (coercive) estimates for solutions of linearized problems permit us to obtain local solutions of nonlinear problems using the contraction mapping principle. The solutions we obtain do not loose regularity with time and are at least as smooth as initial data. The mapping of the unknown domain with a free boundary onto a given domain is made in a different way in the parabolic and in the Navier–Stokes case. In the first case the free boundary Γt is given by an equation x = ξ + ρ(ξ, t) N (ξ), where ξ is an arbitrary point of the given surface Γ0 , N (ξ) is a smooth non-tangential vector field on Γ0 and ρ(ξ, t) is a certain (unknown) function, whereas in the hydrodynamical problems this mapping is defined by relationship between the Eulerian and the Lagrangean coordinates. A central role in our arguments is played by analysis of linearized problems. In both cases they are non-standard, and the proof of coercive estimates relays on the study of corresponding non-standard model problems in the half-space. We are able to write solutions of these problems in the form of

L. Ambrosio et al.: LNM 1812, P. Colli and J.F. Rodrigues (Eds.), pp. 123–175, 2003. c Springer-Verlag Berlin Heidelberg 2003


124

Vsevolod A. Solonnikov

sums of potentials and prove the pointwise estimates of their kernels which makes it possible to obtain necessary estimates of solutions. Major part of the results presented here is published in [7, 8, 35, 40]. These notes have been developed during a visit to the Centro de Matem´ atica e Aplica¸c˜ oes Fundamentais at the University of Lisbon, supported by a grant of the Portuguese Funda¸c˜ao para a Ciˆencia e Tecnologia.

Contents Part I – Parabolic free boundary problems 1. Introduction 2. The model problem 3. Transformation of the problem (1) 4. Proof of Theorem 3.1 Part II – Evolution free boundary problems for the Navier–Stokes equation 5. Introduction 6. Linear and model problems 7. Lagrangean coordinates and local existence theorems 8. Proof of Theorem 5.3 9. Scheme of the proof of Theorem 5.4 References

Part I – Parabolic free boundary problems 1 Introduction Let Ω ⊂ Rn , n ≥ 2, be a bounded domain with the boundary S = ∂Ω divided into two subdomains, Ω1 and Ω2 , by a closed surface Γ which is bounded away from S, and let ∂Ω2 = Γ , ∂Ω1 = S ∪ Γ . One phase Stefan problem consists in the determination of one parameter family of surfaces Γ (t) such that Γ (0) = Γ and of the function u(x, t) given in the domain Ω1 (t) with the boundary ∂Ω1 (t) = Γ (t) ∪ S and satisfying the equations

Lectures on Evolution Free Boundary Problems: Classical Solutions

∂u − a ∆u = 0 , ∂t u|t=0 = u0 (x) , u|x∈S = b(x, t) , u|x∈Γ (t) = 0 ,

125

x ∈ Ω1 (t) , x ∈ Ω1 (0) = Ω1 , (1)  ∂u  Vn = −c0  ∂n 

, x∈Γ (t)

where c0 , a are positive constants and Vn is the velocity of the evolution of Γ (t) in the direction of interior normal n. It is natural to assume that b(x, t) > 0 and u0 (x) > 0 inside Ω1 . We also formulate the closely related two-phase Stefan problem and the Muskat–Verigin problem considered in [2, 3, 6, 17, 22, 26, 27]. In the two-phase Stefan problem it is required to find Γ (t) and the functions x ∈ Ωi (t), i = 1, 2 ,

ui (x, t) ,

(Ω2 (t) = Ω − Ω1 (t)) ,

satisfying the equations ∂ui − ai ∆u = 0 , ∂t ui |t=0 = u0i (x) ,

x ∈ Ωi (t), i = 1, 2 , x ∈ Ωi (0) = Ωi ,

u1 |x∈S = b(x, t) , ui |x∈Γ (t) = 0 ,

 ∂u1 ∂u2  −λ2 Vn = −c0 λ1  ∂n ∂n 

(2) # ≡−c0 x∈Γ (t)

∂u λ ∂n

$    

, x∈Γ (t)

where ai , λi , c0 are positive constants. Finally, the Verigin problem consists in the determination of Γ (t), u1 (x, t), u2 (x, t), such that ∂ui − ai ∆u = 0 , ∂t ui |t=0 = u0i (x) , [u]|x∈Γ (t) = 0 ,

x ∈ Ωi (t), i = 1, 2 , #

x ∈ Ωi (0) ≡ Ωi , u1 |x∈S = b(x, t) ,  $ ∂u1  ∂u  λ = 0, Vn = −c0 λ1   ∂n  ∂n  x∈Γ (t)

(3) . x∈Γ (t)

It will be shown that the problem (1) is uniquely solvable in a certain finite time interval and that the solution belongs to a certain H¨ older space of functions. We recall the definition of these spaces. Let  be an arbitrary

126

Vsevolod A. Solonnikov

positive non-integral number and let Ω be a domain in Rn . By C  (Ω) we mean the set of functions defined in Ω and having a finite norm

() () sup |Dj u(x)| + [u]Ω , |u|C  (Ω) ≡ |u|Ω = 0≤|j|< ()

[u]Ω =

Ω

    sup |x − y|[]− Dxj u(x) − Dyj u(y) ,

|j|=[]

x,y∈Ω

where [] is the integral part of , j = (j1 , ..., jn ), jk are non-negative integers, |j| = j1 + · · · + jn , Dj u(x) =

∂ |j| u(x) j ∂x11 ···∂xjnn

()

. The expression [u]Ω is referred to ()

as a principal part of the H¨ older norm |u|Ω , and separate terms in this expression are H¨ older constants of the derivatives Dj u(x). The boundedness of these constants Mj means that all the derivatives Dj u(x), |j| = [], satisfy the H¨ older condition    j  D u(x) − Dj u(y) ≤ Mj |x − y|−[] , where Mj = sup |x − y|−+[] |Dj u(x) − Dj u(y)|. Further, by anisotropic H¨ older space C ,/2 (QT ), QT = Ω × (0, T ), we mean the set of functions u(x, t), x ∈ Ω, t ∈ (0, T ), with a finite norm (,/2)

|u|QT

()

(/2)

= sup |u(·, t)|Ω + sup |u(x, ·)|(0,T ) . x∈Ω

t∈(0,T )

It is known that if u ∈ C ,/2 (QT ) and |j| + 2 k < , then Dtk Dxj u ∈ 1 C −2k−|j|, 2 (−2k−|j|) (QT ). Hence, a widely used norm

(−2k−|j|, 12 (−2k−|j|)) (,/2) = |Dxj Dtk u|QT uQT 0≤|j|+2k< (,/2)

is equivalent to |u|QT

(,/2)

[u]QT

(,/2)

. By [u]QT

we mean ()

(/2)

= sup [u(·, t)]Ω1 + sup [u(x, ·)](0,T ) . t∈(0,T )

x∈Ω

The space C ,/2 (Γ×(0, T )) is defined in a standard way, i.e. with the help of local maps and of the partition of unity on Γ . Finally, one can introduce the space C ,/2 (QT ) in a non-cylindrical domain QT = {x ∈ Ω(t), t ∈ (0, T )} by mapping this domain onto a cylinder Ω×(0, T ) by an appropriate coordinate transformation. An example of such a transformation is given below. The theorem on the solvability of one-phase Stefan problem may be stated as follows. Theorem 1.1. Assume that the surfaces S and Γ belong to the class C 2+α , α ∈ (0, 1), and that b(x, t) > 0, b ∈ C 2+α, 1+α/2 (S × (0, T )). Further, let u0 ∈ C 2+α (Ω) satisfy the compatibility conditions

Lectures on Evolution Free Boundary Problems: Classical Solutions

u0 (x)|x∈Γ = 0 ,

u0 (x)|x∈S = b(x, 0) ,

bt (x, 0) − ∆u0 (x)|x∈S = 0 ,

2   ∂u0  = 0, ∆u0 − c0   ∂n

127

(4) (5)



(6)

x∈Γ

and the conditions u0 (x) ≥ 0,  ∂u0   ≥ d0 > 0 . ∂n  Γ

Then the problem (1) has a unique solution (Γ (t), u(x, t)) defined for t ∈ (0, T0 ), T0 ∈ (0, T ) and such that Γ (t) belongs to the class C 2+α , ut (·, t) ∈ C α (Ω(t)), u(·, t) ∈ C 2+α (Ω1 (t)) for all t ∈ (0, T0 ), and   (α) (2+α) (2+α) (2+α, 1+α/2) . sup |ut (·, t)|Ω1 (t) + sup |u(·, t)|Ω1 (t) ≤ c |u0 |Ω1 + |b|S×(0,T0 ) t
t
In fact, the solution possesses some additional properties (see Theorem 3.1 below). Similar theorems hold for the two-phase Stefan problem, the Verigin problem and some other problems which may be reduced to the pseudodifferential equation A v = h on Γ ×(0, T ) where A is a pseudodifferential operator with the principal symbol s + φ(ξ, s) (see remark in Section 2). Our methods may be also applicable to a larger class of problems, for instance, problems containing mean curvature in the boundary conditions or quasistationary problems (see [10, 11, 13, 14, 21, 27, 29, 47] and the bibliography in these papers) but we do not discuss it here. The Dirichlet condition u|s = b in (1) can be replaced by another bound1+α ∂u + β u|s = b(x, t), b ∈ C 1+α, 2 (S ×(0, T )). In ary condition, for instance, ∂n this case, condition (5) has no sense. The compatibility conditions (4)–(6) are necessary for the existence of the 2 u continuous in the closed domain solution with the derivatives ut and ∂x∂i ∂x j QT . In particular, (6) is a consequence of the equation ∆u0 = ut |t=0, x∈Γ and of the boundary conditions on Γ . Indeed, differentiation of u(x, t)|x∈Γ (t) = 0 with respect to t leads to 2 ∂u ∂u ut = −∇u · xt = − Vn = c0 , ∂n ∂n

128

Vsevolod A. Solonnikov

hence,

ut |t=0, x∈Γ = ∆u0 = c0

∂u0 ∂n

2 .

The restrictions (5), (6) on u0 (x) have no physical meaning and they can be avoided, if the solution is sought in larger spaces, for instance in weighted ,/2 H¨ older spaces Cs (QT ), s ∈ [0, ], or in the Sobolev spaces. The norm in ,/2 Cs (QT ) is given by (,/2)

(s,s/2)

|u|s,QT = |u|QT

+



sup t

s<2k+|j|<

t
2k+|j|−s 2

|Dtk Dxj u(x, t)|+sup t t
−s 2

(,/2)

[u]Q


t

,

Qt = Ω ×(t/2, t) . ,/2

It is clear that C (QT ) = C ,/2 (QT ) but if s < , then the derivatives ,/2 k j Dt Dx u(x, t) of the elements of Cs (QT ) with s < 2k + |j| <  may have singularities for t = 0. Therefore the equation ut − ∆u|t=0 = 0 for the ele2+α, 1+α/2 ments of Cs (QT ) with s ∈ (1, 2) has no sense, and the first order compatibility conditions (5), (6) are not necessary. Theorems on the solvability of the two-phase Stefan problem and of the Verigin problem in usual and weighted H¨ older spaces may be found in [6, 8]. Different variants of Lp -theory for parabolic free boundary problems are presented in [14, 43]. Further results on the classical solutions approach can also be found in [29], where the combustion free boundary problem can be obtained as a limit case of the Stefan problem as well as an application to a supercondutivity model (see also [46]).

2 The model problem An important role in the proof of Theorem 1.1 is played by the estimates of the solution of the following model problem in the half-space Rn+ = {xn > 0} vt − a ∆v = f (x, t) ,

x ∈ Rn+ , t > 0 ,

v|t=0 = 0 , vt + b · ∇v|xn =0 = g(x , t) ,

(7) x = (x1 , ..., xn−1 ) ∈ Rn−1 .

Here a = const > 0 and b = (b1 , ..., bn ) is a constant vector. This problem does not belong to the class of parabolic initial-boundary value problems, if n > 2 ∂ (because the operator on the boundary ∂t + b · ∇ does not satisfy a standard “complementing” condition which guarantees the solvability). Nevertheless, if bn < 0, the problem (7) is well posed.

Lectures on Evolution Free Boundary Problems: Classical Solutions

129

Theorem 2.1. Assume that bn < 0,

f ∈ C α, α/2 (Rn+ ×(0, T )),

and f (x, 0) = 0 ,

g ∈ C 1+α,

1+α 2

(Rn−1 ×(0, T ))

g(x , 0) = 0 .

Then the problem (7) has a unique solution v ∈ C 2+α, 1+α/2 (Rn+×(0, T )) with 1+α vt ∈ C 1+α, 2 (Rn ×(0, T )), and (2+α, 1+α/2)

[v]Rn ×(0,T ) +

  (1+α, 1+α (1+α, 1+α (α,α/2) 2 ) 2 ) , (8) + [vt |xn =0 ]Rn−1 ×(0,T ≤ c [f ] + [g] n n−1 R ×(0,T ) ) R ×(0,T ) +

+

with the constant independent of T . Proof. We construct the solution in the form v(x, t) = v1 (x, t) + u(x, t), where v1 (x, t) and u(x, t) are solutions of the problems v1t − a ∆v1 = f (x, t) , v1 |t=0 = 0 , and

x ∈ Rn+ ,

v1 |xn =0 = 0

ut − a ∆u = 0, x ∈ Rn+ ,

u|t=0 = 0 ,

ut + b · ∇u|xn =0 = g − b · ∇v1 |xn =0 ≡ h(x , t) ,

(9)

respectively. Both v1 and u can be written explicitly in the form of potentials, namely,
t
  Γ (x − y, t − τ ) − Γ (x − y ∗ , t − τ ) f (y, τ ) dy dτ , v1 (x, t) = 0

Rn +

where y ∗ = (y  , −yn ) and Γ (x, t) is a fundamental solution of the heat equation |x|2 1 −4at e , t>0, Γ (x, t) = (4 π a t)n/2 Γ (x, 0) = 0 , and


t
u(x, t) = 0

where

Rn

t<0,

G(x − y  , xn , t − τ ) h(y  , τ ) dy  dτ,

(10)

130

Vsevolod A. Solonnikov




t

G(x, t) = −2 a 0

∂Γ (x − b λ, t − λ) dλ ∂xn

(see [7]). It is well known that (2+α, 1+ α )

(α,α/2)

[v1 ]Rn ×(0,T )2 ≤ c [f ]Rn ×(0,T ) , +

hence,

(11)

+

(1+α, 1+α )

(1+α, 1+α )

(α,α/2)

2 2 [h]Rn−1 ×(0,T ) ≤ [g]Rn−1 ×(0,T ) + c[f ]Rn ×(0,T ) . +

Let us consider the potential (10). First of all, we estimate the derivatives of the kernel G(x, t). We have
|Dxj G(x, t)|

≤ c

t

(t − λ)−

n+1+|j| 2

e−c

|x−b λ|2 t−λ

dλ

0

− n+1+|j|
t/2 2 |x|2 +λ2 t e−c t−λ dλ ≤ c 2 0


t |x|2 +t2 /4 n+1+|j| −c − t−λ 2 + (t − λ) e dλ t/2



≤ c t− which implies




t

n+|j| 2

e−

c|x|2 t

+ (x2 + t2 )−

|Dxj G(x , xn , t − τ )| dτ ≤

0

n+|j|−1 2

 ,

c |x|n+|j|−2

and, as a consequence, 
t 
 ∂     dτ  G(x − y , xn , t − τ ) dy  ≤ c   |x
−y
|≤a ∂xj 0

In an analogous way it can be proved (see [8]) that
|Dtk G(x , xn , t)| dx ≤ c t−1/2−k ,

k = 0, 1, 2 .

Rn−1

(2+α, 1+α/2)

Let us pass to the estimate of [u]Rn ×(0,T ) +

It can be represented in the form

(13)

∀a > 0 .

j = 1, ..., n−1,

∂ 2 u(x, t) ∂ 2 u(z, t) − , ∂xj ∂xk ∂zj ∂zk

(12)

. Consider the difference

n j, k = 1, ..., n − 1, x, z ∈ R+ .

(14)

Lectures on Evolution Free Boundary Problems: Classical Solutions

131

∂ 2 u(x, t) ∂ 2 u(z, t) − ∂xj ∂xk ∂zj ∂zk

# $ ∂G(x − y  , xn , t − τ ) ∂h(y  , τ ) ∂h(x , τ ) dy  = dτ − ∂xj ∂yk ∂xk |x
−y
|≤2ρ 0 # $
t
∂G(z  − y  , zn , t − τ ) ∂h(y  , τ ) ∂h(z  , τ ) dy  − dτ − ∂z ∂y ∂z

j k k |x −y |≤2ρ 0 $

t# ∂G(x −y  , xn , t−τ ) ∂h(x , τ ) ∂h(z  , τ ) dτ + − dy  ∂x ∂z ∂x

j j j |x −y |≤2ρ 0 # $
t
∂G(x −y  , xn , t−τ ) ∂G(z  −y  , zn , t−τ ) + dτ − ∂xj ∂zj |x
−y
|≥2ρ 0 # $ ∂h(y  , τ ) ∂h(z  , τ ) × dy  , − ∂yk ∂zk


t




where ρ = 2 |x − z|. Using inequalities (12), (13), we obtain    ∂ 2 u(x, t) ∂ 2 u(z, t)    −    ∂xj ∂xk ∂zj ∂zk 
|x −y  |α (α) ≤ c sup[hyk (·, τ )]R2 dy    n−1 τ


|z  −y  |α |z  −y  |α  α  + dy + ρ + ρ dy   n−1   n |x
−y
|≤2ρ |z −y | |x
−y
|≥2ρ |z −y | (α)

≤ c ρα sup[hyk (·, τ )]R2 . τ
Similarly, the representation formula ∂u(x, t + r) ∂u(x, t) − ∂t ∂t
t+r
& ∂G(x − y  , xn , t + r − τ ) %  h(y , τ ) − h(y  , t + r) dy  dτ = ∂t Rn−1 −∞
t
& ∂G(x − y  , xn , t − τ ) %  h(y , τ ) − h(y  , t) dy  − dτ ∂t Rn−1 −∞

(15)

132

Vsevolod A. Solonnikov




=




& ∂G(x − y  , xn , t + r − τ ) %  h(y , τ ) − h(y  , t + r) dy  ∂t Rn−1 t−r
t
& ∂G(x − y  , xn , t − τ ) %  h(y , τ ) − h(y  , t) dy  − dτ ∂t Rn−1 t−r
% & h(y  , t + r) − h(y  , t) G(x − y  , xn , r) dy  + t+r

dτ

Rn−1




#




∂G(x − y  , xn , t + r − τ ) ∂G(x − y  , xn , t − τ ) − + dτ ∂t ∂t Rn−1 −∞ % & × h(y  , τ ) − h(y  , t + r) dy  , t−r

$

where h(y  , τ ) = 0 for τ < 0, implies    ∂u(x, t + r) ∂u(x, t)    −    ∂t ∂t  ≤ c sup [h(y Rn−1




t

+ t−r



( 1+α 2 ) , ·)](0,t+r)




t+r

dτ (t+r − τ )1−α/2

· t−r

dτ +rα/2 +r (t−τ )1−α/2




t−r

−∞

dτ (t+r − τ )2−α/2

( 1+α )

2 , ≤ c hα/2 sup [h(y  , ·)](0,t+r)

Rn−1

i.e.,

( 1+α )

(α/2)

2 sup[ut (·, t)](0,T ) ≤ c sup [h(·, t)](0,T ) .

Rn +

(16)

Rn−1

Finally, we estimate the H¨older constant   # $  2  ∂2u ∂ 2 u(x , xn , t)  −α  ∂ u(x , xn + r, t) ≡ sup sup r  − Mn .   ∂x2n ∂x2n ∂x2n r>0 Rn + ×(0,T ) Since uxn xn = a−1 ut −

"n−1 j=1

uxj xj , we have

Mn [uxn xn ] ≤ a−1 Mn [ut ] +

n−1

Mn [uxj xj ] .

j=1

The last term is already estimated, and the first term we estimate by interpolation inequality

Lectures on Evolution Free Boundary Problems: Classical Solutions

133

(α)

Mn [ut ] ≤ ε Mn [uxn xn ] + c(ε) sup[ut (x, ·)](0,T ) , Rn +

(see [30]) with ε  1. This yields, by virtue of (15) and (16), (1+α, 1+α )

2 Mn [uxn xn ] ≤ c [h]Rn−1 ×(0,T ) .

This estimate, together with (15) and (16), implies (2+α, 1+α/2)

[u]Rn ×(0,T ) +

(1+α,

1+α

)

2 ≤ c [h]Rn−1 ×(0,T ) .

(17)

Inequality (8) follows from (11), (17) and from the equation vt + b · ∇v|xn =0 = g. Now, it can be easily shown that the function v(x, t) is bounded; indeed,
t
t  T 1+α (2+α,1+α/2)   |v(x, t)| ≤ [v] n |vτ (x, τ )|dτ ≤ . vτ (x, τ )−vτ (x, 0)dτ ≤ 1+α R+ ×(0,T ) 0 0 2

∂v v and ∂x∂i ∂x are also bounded for arbitrary t ∈ Hence, the derivatives ∂x i m 2+α, 1+α/2 (Rn+ ×(0, T )). (0, T ), T < ∞, and v ∈ C Let us show that the solution of the problem (7) constructed above is unique in the class C 2+α, 1+α/2 (Rn+ ×(0, T )), i.e., that arbitrary solution u0 ∈ C 2+α, 1+α/2 (Rn+ × (0, T )) of a homogeneous problem vanishes. At first we assume additionally that u0 and its derivatives belong to L2 (Rn+ × (0, T )). Then there holds the energy equality
0 = u0 (u0t − a ∆u0 ) dx dτ Rn +


 ∂u 1 d  0 u0 (·, t)2L2 (Rn ) + a ∇u0 (·, t)2L2 (Rn ) + a u0 dx =  + + 2 dt ∂x n−1 3 R x3 =0

1 d a u0 (·, t)2L2 (Rn ) − = u0 |x3 =0 2L2 (Rn−1 ) + a ∇u0 (·, t)2L2 (Rn ) , + + 2 dt bn from which it follows that u0 = 0. Now, we assume the regularity C 2+α, 1+α/2 (Rn+ ×(0, T )) for u0 and we introduce the family of cut-off funcx tions ζR (x) = ζ( R ), R  1, where ζ ∈ C0∞ (Rn+ ), ζ(z) = 1 for |z| < 12 , ζ(z) = 0 for |z| > 1. It is clear that the function uR 0 = u0 ζR is a solution of the problem (7) with f = −2 a ∇ζR · ∇u0 − a u0 ∇2 ζR , g = u0 b · ∇ζR . Since, n uR 0 ∈ L2 (R+ ), ∀ t ∈ (0, T ), it is expressed in terms of f and g as indicated above (because it can be easily verified that the solution constructed above belongs to L2 (Rn+ ), if f and g have a compact support). Hence (2+α, 1+α/2)

|uR 0 |Rn ×(0,T ) +

(1+α, 1+α )

≤ |f |Rn ×(0,T ) + |g|Rn ×(0,T2 ) ≤ c R−1 (α,α/2) +

+

n and, as a consequence, u0 = uR 0 (x, t) → 0 as R → ∞ for arbitrary x ∈ R+ , t ∈ (0, T ), and the theorem is proved.

134

Vsevolod A. Solonnikov

Remark 2.1. Let us come back to the problem (9). Making the Fourier– Laplace transformation with respect to (x , t), we can write the solution in the form √s 2  h(ξ, s) − a +ξ xn s e , u (ξ, xn , t) = s + α + ξ2 where u ,  h are transformed functions. Similar formulas hold for the solutions of model problemsarising in the s 2 study of free boundary problems (2), (3), but the function a + ξ is replaced with some other homogeneous functions analytical in the domain Re s > −δ ξ 2 , δ  1.

3 Transformation of the problem (1) It is convenient to write the problem (1) as a nonlinear initial-boundary value problem in a given domain Ω. It can be achieved by a special mapping of the domain Ω1 (t) onto Ω. Let N (ξ), ξ ∈ Γ , be a regular (at least of class C 2+α (Γ )) vector field such that N (ξ) · n0 (ξ) ≥ ν0 > 0, where n0 (ξ) is a unit interior (with respect to Ω1 ) normal to Γ . For instance, we can set "M ϕk (ξ) n0 (ξk )  , N (ξ) = "k=1  M  k=1 ϕk (ξ) n0 (ξk ) where {ϕk (ξ)}k=1,...,M is a partition of unity on Γ (and in a certain neighbourhood of Γ ) subordinate to the covering of Γ by the balls |ξ −ξk | < δ  1, ξk ∈ Γ . It is easily seen that the set ( ' N = ξ + N (ξ) λ, 0 < λ < λ0 , ξ ∈ Γ ∪ Ω contains a certain neighbourhood Ω  of Γ and that the equation x = ξ + N (ξ) λ defines the functions ξ(x), λ(x) of the class C 2+α (Ω  ) (we observe that if N (ξ) = n0 (ξ), then ξ(x) and λ(x) belong to C 1+α (Ω  )). Hence, for small t the surface Γ (t) can be defined by the equation λ = ρ(ξ, t), ξ ∈ Γ , i.e., ' ( Γ (t) = x ∈ Rn : x = ξ + N (ξ) ρ(ξ, t), ξ ∈ Γ , where ρ is a certain (unknown) function of the class C 2+α, 1+α/2 (Γ ×(0, T )). To every such function we put into correspondence its extension

Lectures on Evolution Free Boundary Problems: Classical Solutions

135

ρ∗ (ξ, t) = P ρ(ξ, t), where P : C 2+α, 1+α/2 (Γ ×(0, T )) → C 2+α, 1+α/2 (Ω1 ×(0, T )) is a continuous linear operator such that  ∂ρ∗ (ξ, t)  ∗ = 0, ρ∗ (ξ, t)|ξ∈S = 0 , ρ (ξ, t)|ξ∈Γ = ρ(ξ, t) ,  ∂n0  ξ∈Γ

and

|ρ∗ |Ω1 ×(0,t)

(2+α, 1+α/2)

(2+α, 1+α/2)

≤ c |ρ|Γ ×(0,t)

.

We also extend N (ξ) from Γ into Ω1 : first into Ω  by the formula N (x) = N (ξ(x)) and then into Ω1 with the preservation of class C 2+α and in such a way that N (x) = 0 near S. Consider the mapping x = y + N (y) ρ∗ (y, t) ≡ eρ (y, t) ,

y ∈ Ω1 .

For small ρ, this is an invertible mapping of Ω1 onto Ω1 (t), whose Jacobi matrix J = (Jkm )k,m=1,...,n has the elements Jkm = δkm +

∂ρ∗ ∂ ∂Nk ∗ Nk ρ∗ = δkm + Nk + ρ . ∂ym ∂ym ∂ym

By∗J km we mean the elements of J −1 . They are algebraic functions of ρ∗ and ∂ρ ∂ym . The normal n0 (y) to Γ and the normal n(eρ (y, t)) to Γ (t) are related to each other as follows J −T n0 , n = −T |J n0 | J −T = (J −1 )T . It is easily seen that (1) is equivalent to the problem of determination of ρ(y, t), y ∈ Γ , and of U (y, t) = u(eρ (y, t), t), y ∈ Ω1 , such that Lρ U = 0 , y ∈ Ω, t ∈ (0, T ) , U |t=0 = u0 (y) , U |y∈S = b(y, t) ,

ρ|t=0 = 0 , U |y∈Γ = 0 ,

ρt (N · J −T n0 ) + c0 n0 · A ∇U |y∈Γ = 0 , where A = J

−1

J

−T

Amp

is the matrix with the elements

n

∂ ∂ mk pk = J J , ∇= , ..., ∂y1 ∂yn k=1

(18)

(19)

136

Vsevolod A. Solonnikov

and



∂U −(N · J −T ∇U ) ρ∗t −a Lρ U = ∂t

n



∂2U ∂J pk ∂U Amp + J mk ∂ym∂yp ∂ym ∂yp m,p=1

k,m,p=1

 N ·J −T n0  (we have used the equation Vn = ∂x ∂t · n|Γ (t) = ρt |J −T n0 | Γ ). Since U |Γ = 0, the condition (19) is equivalent to  ∂U  (20) ρt + c0 Sρ  = 0, ∂n0  Γ

where

Sρ = (N · J −T n0 )−1 (n0 · A n0 ) .

Further, we introduce auxiliary functions ρ0 (ξ, t), ξ ∈ Γ , and U0 (y, t), y ∈ Ω, satisfying the same initial conditions as ρ and U , namely, ρ0 |t=0 = 0 ,

ρ0t |t=0 = −c0 (N · n0 )−1

U0 |t=0 = u0 ,

∂u0 , ∂n0

U0t |t=0 = (N · ∇u0 ) ρ∗0t + a ∆u0 .

We construct ρ0 making use of the following proposition (cf. [8]) Proposition 3.1. For arbitrary f0 ∈ C 1+α (Γ ) there exists ρ(0) ≡ R f0 ∈ C 2+α, 1+α/2 (Γ ×d(0, T )) (0)

such that ρt ∈ C 1+α,

(Γ ×d(0, T )) ,

(0)

ρ(0) |t=0 = 0 ,

ρt |t=0 = f0 ,

and (2+α, 1+α/2)

|ρ(0) |Γ ×(0,T )

1+α 2

(0) (1+α,

1+α

+ |ρt |Γ ×(0,T )2

)

(1+α)

≤ c |f0 |Γ

with the constant independent of T . The operator R is linear. We set ρ0 = R(−c0 (N · n0 )−1

∂u0 ∂n0 )

and define U0 as a solution of the problem

Lρ0 U0 = 0 ,

y ∈ Ω1 ,

U0 |t=0 = u0 (y) , U0 |y∈S = b ,

U0 |y∈Γ = 0 .

In virtue of (4)–(6), necessary compatibility conditions are satisfied, and this problem has a unique solution U0 ∈ C 2+α, 1+α/2 (Ω1 ×(0, T )) satisfying the inequality

Lectures on Evolution Free Boundary Problems: Classical Solutions (2+α, 1+α/2)

|U0 |Ω1 ×(0,T )



(2+α, 1+α)

≤ c |b|S×(0,T )

(2+α)

+ |u0 |Ω

137

 .

Further, we set r = ρ − ρ0 , V = U − U0 , δ Sρ0 =

d Sρ +λr |λ=0 , dλ 0

δ Lρ0 = Lρ0 +λr |λ=0 ,

and we write (18), (20) as the problem of determination of r and V such that Lρ0 V + δ Lρ0 U0 = F(r, V ) , V |t=0 = 0 ,

r|t=0 = 0 ,

V |y∈S = 0 ,

V |y∈Γ = 0 ,

rt + c0 Sρ0

∂V ∂U0 + δ Sρ0 ∂n ∂n0

y∈Ω , (21)

= G(r, V ) + ϕ(y, t) ,

y∈Γ ,

where F(r, V ) = −(Lρ − Lρ0 − δ Lρ0 ) U0 − (Lρ0 − Lρ0 ) V , G(r, V ) = −c0 (Sρ − Sρ0 − δ Sρ0 ) ϕ = −ρ0t − c0 Sρ0

∂U0 ∂V − c0 (Sρ − Sρ0 ) , ∂n0 ∂n0

(22)

∂U0 . ∂n0

It is clear that ϕ(y, 0) = 0. The expressions δ Sρ0 and δ Lρ0 depend on δ J0−1 = −J0−1 (δ J0 ) J0−1 = −J0−1 (∇N r∗ )T J0−1 and can be easily calculated. This leads to the following form of equations (21) Lρ0 V − (N · J0−T ∇U0 ) Lρ0 r∗ + m(r∗ , V ) = F(r, V ) , V |t=0 = 0 ,

r|t=0 = 0 ,

V |y∈S = 0 ,

V |y∈Γ = 0 ,

rt + c0 Sρ0

y ∈ Ω1 , (23)

∂V + h · ∇Γ r + h0 r|y∈Γ = G(r, V ) + ϕ(y, t) . ∂n0

Here h = (h1 , ..., hm ) ∈ C 1+α, is a gradient on Γ , r∗ = Pr, Lρ0 w = wt − a

1+α 2

n

i,j=1

(Γ ×(0, T )), h0 ∈ C 1+α,

(0)

Aij

∂2w , ∂yi ∂yj

(0)

1+α 2

(Γ ×(0, T )), ∇Γ

Aij = Aij |ρ=ρ0 ,

138

Vsevolod A. Solonnikov

and m(r∗ , V ) is a linear first order operator of the type m(r∗ , V ) = η  · ∇r∗ + η  · ∇V + η0 r∗ + η0 V , with the coefficients belonging to C α,α/2 (Ω1 ×(0, T )). Theorem on the solvability of the problem (23) can be stated as follows. Theorem 3.1. Under the hypotheses of Theorem 1.1, the problem (23) has a unique solution (V, r) such that V ∈ C 2+α, 1+α/2 (Ω1 × (0, T0 )), r ∈ 1+α C 2+α, 1+α/2 (Γ ×(0, T0 )), rt ∈ C 1+α, 2 (Γ ×(0, T0 )), defined on a certain time interval (0, T0 ) and satisfying the inequality (2+α, 1+α/2)

|V |Ω1 ×(0,T0 )

(2+α, 1+α/2)

+ |r|Γ ×(0,T0 )

(1+α, 1+α )

+ |rt |Γ ×(0,T02)

(1+α, 1+α )

≤ c |ϕ|Γ ×(0,T02)

.

(24)

Theorem 1.1 follows from Theorem 3.1.

4 Proof of Theorem 3.1 Assume that ρ0 (y, t), y ∈ Γ , t ∈ (0, T ), is so small that the transformation X = eρ0 (y, t) is invertible, N · J0−1 n0 and |J0−1 n0 | are strictly positive, and consider a linear problem Lρ0 V − q(y, t) Lρ0 r∗ + m(r∗ , V ) = f (y, t) , V |t=0 = 0 ,

V |y∈S = 0 ,

y ∈ Ω1 , t ∈ (0, T ) ,

V |y∈Γ = 0 ,

(25)

∂V rt + b + h · ∇Γ r + h0 r|y∈Γ = g(y, t) , ∂n where q ∈ C α,α/2 (Ω1 ×(0, T )), b ∈ C 1+α,

1+α 2

(Γ ×(0, T )).

Theorem 4.1. If b ≥ b0 > 0, q ≥ q0 > 0 then for arbitrary f ∈ C α,α/2 (Ω1 × 1+α (0, T )), g ∈ C 1+α, 2 (Γ ×(0, T )) satisfying the conditions f (y, 0) = 0 ,

g(y, 0) = 0 ,

the problem (25) has a unique solution (V, r) such that V ∈ C 2+α, 1+α/2 (Ω1× 1+α (0, T )), r ∈ C 2+α, 1+α/2 (Γ ×(0, T )), rt ∈ C 1+α, 2 (Γ ×(0, T )) and (2+α, 1+α/2)

(2+α, 1+α/2)

(1+α, 1+α )

Y(T ) ≡ |V |Ω1 ×(0,T ) + |r|Γ ×(0,T ) + |rt |Γ ×(0,T 2)   (1+α, 1+α ) (α,α/2) ≡ c N (T ) . ≤ c |f |Ω1 ×(0,T ) + |g|Γ ×(0,T 2)

(26)

This theorem is proved in the same way as Theorem 4.2 in [8], and we restrict ourselves with sketching main ideas of the proof of the estimate (26).

Lectures on Evolution Free Boundary Problems: Classical Solutions

139

We use classical Schauder’s method and estimate at first the function r(y, t), y ∈ Γ . Let us consider a neighbourhood of arbitrary point y0 ∈ Γ . Without restriction of generality we can assume that y0 = 0 and the yn -axis is directed along the interior normal n(y0 ). Let yn = F (y  ) ,

y  = (y1 , ..., yn−1 ) ∈ Kd ≡ {|y  | < d}

be equation of Γ near the origin; clearly, F ∈ C 2+α (Kd ) and ∇ F (0) = (Fy1 (0), ..., Fyn−1 (0)) = 0 .

F (0) = 0 ,

We make the change of variables z = y ,

zn = yn − F (y  ),

which “flattens” Γ near the origin. In the new variables (z, t) (25) takes the form  ρ V − q(z, t) L  ρ (r∗ ) + m(r L  ∗ , V ) = f (z, t) , 0 0 V |t=0 = 0 ,

V |zn =0 = 0 ,

(27)

∂V rt + b + h · ∇ r + h0 r|zn =0 = g(z  , t) , ∂zn where

n 2

ρ = ∂ − a ij ∂ , L A 0 ∂t ∂zi ∂zj i,j=1

ij (0, t) = Aij (0, t) , A

ij (0, 0) = Aij (0, 0) = δij , A

and m  is an operator of the same type as m. We introduce a new unknown function W (z, t) = V (z, t) − q(0, 0) r∗ (z, t) and obtain, instead of (27),      ρ W = q(z, t) − q(0, 0) L ρ r∗ − m L  r∗ , W + q(0, 0) r∗ + f (z, t) , 0 0 W |t=0 = 0 ,

  h − b(z, t) q(0, 0) ∇ F · ∇ W + h0 W |zn =0 Wt − q(0, 0) b(z, t) Wzn +  = −q(0, 0) g  (z  , t) . (28)

Finally, we multiply (28) by the cut-off function ζλ (z), λ < d (ζλ is defined above) and we set

140

Vsevolod A. Solonnikov

w(z, t) = W (z, t) ζλ (z)

if zn > 0, |z| < λ ,

w(z, t) = 0

if zn > 0, |z| ≥ λ .

This function can be considered as a solution of the problem ∂w − a ∆w = f1 (z, t) , ∂t

w|t=0 = 0 ,

wt − β(0, 0) · ∇w|zn =0 = g1 (z, t) , where n 

   ij (z, t) − A ij (0, 0) wz z − Lρ ζλ − ζλ Lρ W A i j 0 0

f1 (z, t) = a

i,j=1

     r∗ , W + q(0, 0)r∗ + f ζλ , + q(z, t) − q(0, 0) ζλ Lρ0 r∗ − ζλ m g1 (z  , t) =



 β(z, t) − β(0, 0) · ∇w + W β(z, t) · ∇ζλ

− h0 w − ζλ q(0, 0) g(z, t)

if |z| < λ ,

f1 (z, t) = 0, g(z  , t) = 0 if |z| ≥ λ ,   h, q(0, 0) b(z, t) . β(z, t) = b(z, t) q(0, 0) ∇ F −  In virtue of Theorem 2.1, we have (2+α, 1+α/2)

[w]Rn ×(0,τ ) +

(1+α, 1+α/2)

+ [wt ]Rn−1 ×(0,τ )

  (1+α, 1+α (α,α/2) 2 ) ≤ c [f1 ]Rn ×(0,τ ) + [g1 ]Rn−1 ×(0,τ ) , +

for arbitrary τ < T . Hence, (1+α, 1+α )

(2+α, 1+α/2)

2 + |wt |Rn−1 ×(0,τ )

 α/2 τ + λα ) X(τ ) + Y(τ ) + c(λ) N (τ ) , λ2

X(τ ) ≡ |w|Rn ×(0,τ ) + α

≤ c (τ 2

where Y(τ ) and N (τ ) are defined in (26). Since the point y0 ∈ Γ is arbitrary, this inequality implies, for small τ and λ, (2+α, 1+α/2) |r|Γ ×(0,τ )

+

(1+α, 1+α ) |rt |Γ ×(0,τ )2



τ ≤ c 2 λ

α/2 Y(τ ) + c(λ) N (τ ) .

Now, we consider V as a solution of a parabolic problem Lρ0 V + m(r∗ , V ) = f − q Lρ0 r∗ , V |t=0 = 0 ,

V |y∈S = 0 ,

V |y∈Γ = 0 ,

Lectures on Evolution Free Boundary Problems: Classical Solutions

141

with a given r(y, t) and obtain 

τ Y(τ ) ≤ c 2 λ

α/2 Y(τ ) + c(λ) N (τ ) .

If τ /λ2 is small, this inequality yields (26). Hence, (26) is proved in a small time interval (0, τ ). As shown in [8], this result can be easily generalized to the case of arbitrary T > 0. Theorem 3.1 can be now deduced from the contraction mapping principle. We write the problem (23) in the form (V, r) = M(F, G + ϕ) ≡ M(F, G) + (V0 , r0 ) ,

(29)

where (V0 , r0 ) = M(0, ϕ), and M is a linear operator which makes correspond to (f, g) the solution of a linear problem Lρ0 V − (N · J0−T ∇U0 ) Lρ0 r∗ + m(r∗ , V ) = f (y, t) , V |t=0 = 0 , rt + c0 Sρ0

r|t=0 = 0 ,

V |y∈S = 0 ,

V |y∈Γ = 0 ,

∂V + h · ∇Γ r + h0 r|y∈Γ = g(y, t) . ∂n

We consider (29) in the space Xα of couples of functions (V, r) with the norm (2+α, 1+α/2)

(V, r)XT = |V |Ω1 ×(0,T )

(2+α, 1+α/2)

+ |r|Γ ×(0,T )

(1+α, 1+α )

+ |rt |Γ ×(0,T 2)

.

It can be easily verified that the functions F(r∗ , V ) and G(r∗ , V ) defined in (22) satisfy the inequalities (1+α, 1+α )

1+α

|F(r∗ , V )|Ω1 ×(0,τ ) + |G(r∗ , V )|Γ ×(0,τ )2 ≤ c τ 2 (V, r)2Xα (τ ) ,  (α,α/2)  (1+α, 1+α 2 )     + G(r∗ , V1 ) − G(r2∗ , V2 ) F(r∗ , V1 ) − F(r2∗ , V2 ) Ω1 ×(0,τ ) Γ ×(0,τ )   α  1+α (1+α, 1+ ) (2+α, 1+α/2) (V1 , r1 )Xτ +(V2 , r2 )Xτ , ≤ c τ 2 |r1 −r2 |Γ ×(0,τ ) +|V1 −V2 |Γ ×(0,τ2) (α,α/2)

(30) if ρ0 , r, r1 , r2 are sufficiently small. Therefore, for small T = T0 , the contraction mapping principle guarantees the existence of a unique solution of the equation (29) in the ball (V, r)XT0 ≤ 2 (V0 , r0 )XT0 . There is no other solution with a finite norm (V, r)XT0 which can be easily deduced from the estimates (26) and (30). Thus, Theorem 3.1 is proved.

142

Vsevolod A. Solonnikov

Part II – Evolution free boundary problems for the Navier–Stokes equation 5 Introduction A typical example of evolving interface is provided by a free surface of a moving liquid or an interface between two different moving liquids. We mention here two evolution free boundary problems for viscous flow studied in the most detailed way. The first of them concerns the motion of a viscous incompressible liquid over a rigid bottom and can be stated as follows. It is necessary to find the domain Ωt = {−b < x3 < η(x , t), x = (x1 , x2 ) ∈ R2 }, i.e., the function η(x , t), the velocity vector field v(x, t) = (v1 (x, t), v2 (x, t), v3 (x, t)) and a scalar pressure p(x, t), x ∈ Ωt , satisfying the Navier–Stokes equations vt + (v · ∇) v − ν ∇2 v + ∇p = 0 , ∇ · v = 0,

x ∈ Ωt , t > 0 ,

and the following initial and boundary conditions v(x, 0) = v0 (x) , x ∈ Ω0 = {−b < x3 < η(x , 0)} η(x , 0) = η0 (x ) , x ∈ R2 , ηt = v3 −

2

ηxj vj ,

x ∈ Γt = {x3 = η(x , t)} ,

(31) (32)

j=1

T (v, p) n = σ H n − g η(x , t) n , v|x3 =b = 0 .

x ∈ Γt ,

(33) (34)

Here g, ν > 0 and σ ≥ 0 are given constants, namely, gravitational constant, coefficient of viscosity and of the surface tension, respectively, n is an exte∂v k rior normal to Γt , T (v, p) = −p I + ν S(v) and S(v) = ( ∂xkj + ∂v ∂xj )j,k=1,2,3 ≡ (Sjk )j,k=1,2,3 are the stress tensor and the doubled rate-of-strain tensor, respectively, 2

ηxj ∂  H = ∂x 1 + ηx21 + ηx22 j j=1 is the doubled mean curvature of Γt ,

3

∂vj ∂ ∂ ∂ ∇= , ∇p = grad p, ∇ · v = , , = divv . ∂x1 ∂x2 ∂x3 ∂x j j=1 The kinematic boundary condition (32) tells that the velocity of the evolution of the free surface Γt in the normal direction Vn = 

ηt 1 + ηx21 + ηx22

Lectures on Evolution Free Boundary Problems: Classical Solutions

143

coincides with vn = v·n. The dynamic boundary condition (33) expresses the balance of forces at the free boundary Γt . Finally, (34) is a standard no-slip condition at the bottom x3 = −b. The problem governing the evolution of an isolated liquid mass consists in the determination of a bounded domain Ωt ⊂ R3 , and of the functions v(x, t) = (v1 , v2 , v3 ) and p(x, t) satisfying the relations vt + (v · ∇) v − ν ∇2 v + ∇p = 0 , v|t=0 = v0 (x) ,

∇ · v = 0, x ∈ Ωt , t > 0 ,

x ∈ Ω0 (Ω0 ≡ Ω is given) ,

(35)

Vn = v · n|Γt .

T (v, ρ) n − σ H n|Γt = 0 ,

Problems (31)–(34) and (35) have been studied in the papers [1, 4, 5, 33– 37]. It has been shown that both of them are uniquely solvable in a certain finite time interval, but if σ > 0 and the initial data are close to the rest state, then the solution exists for all t > 0. The same is true for the solution of the problem (35) with σ = 0, if v0 (x) is small and satisfies additional condition expressing the fact that the total momentum of the liquid equals zero. The solutions were found in anisotropic Sobolev spaces. There have been studied other related problems, such as evolution of isolated mass of viscous compressible liquid or non-steady motion of two liquids (see for instance [12, 44]) but they are not treated here. In what follows we give the main ideas of the proof of the solvability of the problem (35) in the H¨ older spaces of functions (more elementary than in [24]). Our main results may be stated as follows. Theorem 5.1. Let Γ ≡ ∂Ω ∈ C 2+α , α ∈ (0, 1). For arbitrary v0 ∈ C 2+α (Ω) satisfying the compatibility conditions ∇ · v0 = 0 ,

S(v0 ) n0 − n0 (n0 · S(v0 ) n0 )|Γ = 0,

(36)

where n0 is an exterior normal to Γ , problem (35) with σ = 0 has a unique solution defined in a certain finite time interval (0, T0 ) and possessing the following properties: Γt ∈ C 2+α , v(·, t) ∈ C 2+α (Ωt ), vt (·, t) ∈ C α (Ωt ), p(·, t) ∈ C 1+α (Ωt ) for all t ∈ [0, T0 ]. The solution satisfies the inequality (α)

(2+α)

sup |vt (·, t)|Ωt + sup |v(·, t)|Ωt

t
t
(2+α)

≤ c|v0 |Ω

(2+α)

(1+|v0 |Ω

t
(37)

).

(2+α)

The number T0 depends on |v0 |Ω tends to 0.

(1+α)

+ sup |p(·, t)|Ωt

(2+α)

and it tends to infinity as |v0 |Ω

144

Vsevolod A. Solonnikov

Theorem 5.2. Let σ > 0, Γ ∈ C 3+α , α ∈ (0, 1). For arbitrary v0 ∈ C 3+α (Ω) satisfying the conditions (36) the problem (35) has a unique solution Γt ∈ C 2+α , v(·, t) ∈ C 2+α (Ωt ), vt (·, t) ∈ C α (Ωt ), p(·, t) ∈ C 1+α (Ωt ) ∀ t ∈ [0, T0 ], and (α)

(2+α)

(1+α)

sup |vt (·, t)|Ωt + sup |v(·, t)|Ωt + sup |p(·, t)|Ωt t
t
(38)

where H0 is the doubled mean curvature of Γ = ∂Ω. Theorem 5.3. Assume that Γ and v0 (x) satisfy the hypotheses of Theorem 5.1 and that (2+α)

|v0 |Ω

≤ ε 1,
v0 (x) dx = 0, [v0 ×x] dx = 0 .


Ω0

(39) (40)

Ω0

Then the solution of the problem (35) with σ = 0 is defined for all t > 0, Γt ∈ C 2+α for t > 0 and (37) holds for T = ∞, moreover, (α)

(2+α)

|vt (·, t)|Ωt + |v(·, t)|Ωt

(1+α)

+ |p(·, t)|Ωt

≤ c e−bt |v0 |Ω

(2+α)

.

(41)

The formulation of the global existence theorem in the case σ > 0 needs some preparations. We observe that the problem (35) possesses some integrals of motion (both for σ > 0 and for σ = 0), namely,

v(x, t) dx = v0 (x) dx (conservation of momentum) , Ωt

Ω







[v(x, t)×x]dx = Ωt

[v0 (x)×x]dx (conservation of an angular momentum), Ω

|Ωt | = |Ω|

(conservation of mass) .

Without restriction of generality, it can be assumed that

v(x, t) dx = 0 , [v×x] dx = β e3 , e3 = (0, 0, 1) , Ωt

Ωt

because to this case one can always arrive by a passage to a new inertial coordinate system. Moreover, problem (35) has a stationary solution corresponding to the rotation of the liquid as a rigid body about the x3 -axis with the angular momentum β. We denote this solution by v∞ , p∞ , Ω∞ . It is defined by

Lectures on Evolution Free Boundary Problems: Classical Solutions

v∞ = −

β h(x) , I∞

h(x) = (−x2 , x1 , 0) ,

145

β |h(x)|2 + p1 , 2 2 I∞    x , = |x| < R∞ |x|

p∞ (x) = Ω∞

 where p1 = const, I∞ = Ω∞ |h(x)|2 dx and R∞ is a solution of the following differential equation on the unit sphere S1 ⊂ R3 H[R∞ ] +

β2 2 R∞ (y) (y12 + y22 ) + p1 = 0 , 2 2 I∞

|y| = 1 .

(42)

x ), i.e., Here H[R] is the doubled mean curvature of the surface |x| = R( |x|

1 2R 1 ∆S1 R , (43) H[R] = √ + ∇S1 √ · ∇S1 R − √ R g g g

g = R2 + |∇S1 R|2 , ∇S1 and ∆S1 are the gradient and the Laplacean on S1 , respectively. It is shown in [37] that for small β equation (42) has a unique solution R∞ ∈ C ∞ (S1 ) such that |Ω∞ | = |Ω|

and

sup |∇Sj1 R∞ (y)| ≤ c(|j|) |β| .

y∈S1

The condition |Ω∞ | = |Ω| defines the constant p1 in a unique way. Theorem 5.4. Assume that the surface Γ = ∂Ω is given by the equation

x |x| = R ,0 |x| on S1 . If the constant β is small and, in addition, (3+α)    (2+α) |v0 − v∞ |Ω + R(·, 0) − R∞  ≤ ε1  1, S1

then the solution of the problem (35) is defined for t > 0, the surface Γt is given by the equation

x ,t (44) |x| = R |x| and

(2+α)   (1+α)  (3+α)      (α)  + p(·, t)−p∞ (·) + R(·, t)−R∞ (·) |vt (·, t)|Ωt + v(·, t)−v∞  Ωt Ωt S1  (3+α)     (2+α) . ≤ c e−bt |v0 −v∞ |Ω + R(·, 0)−R∞  S1

(45)

146

Vsevolod A. Solonnikov

6 Linear and model problems The proof of Theorems 5.1–5.4 is based on the analysis of two linear problems vt − ν ∇2 v + ∇p = f (x, t), ∇ · v = g(x, t) ,

x ∈ Ω, t > 0 ,

v|t=0 = v0 (x) ,

(46)

T (v, p) n0 |Γ = a(x, t), and vt − ν ∇2 v + ∇p = f (x, t), ∇ · v = g(x, t) , v|t=0 = v0 (x) ,



T (v, p) n0 − σ n0

x ∈ Ω, t > 0 ,


t   v(x, τ ) dτ  = b + n0 B dτ ,  0 0


n0 · ∆Γ

(47)

t

Γ

where Ω is a given bounded domain in R3 , Γ ≡ ∂Ω, n0 is an exterior normal to Γ and ∆Γ is the Laplace–Beltrami operator on Γ . Problem (46) is the second (Neumann) initial-boundary value problem for the Stokes equations studied in the papers [31, 32, 39]. The following result is obtained in [39]. Theorem 6.1. Let Γ ∈ C 2+α , α ∈ (0, 1), v0 ∈ C 2+α (Ω), and let f , g, a be continuous functions such that f ∈ C α (Ω), g ∈ C 1+α (Ω), a ∈ C 1+α (Γ ) for all t ∈ [0, T ], and 0

a ≡ a − n0 (n0 · a) ∈ C 1+α,

1+α 2

(Γ ×(0, T )) .

Assume also that the compatibility conditions ∇ · v0 = g(x, 0) , 0 (S(v0 ) n0 )|Γ

=

0

(48)

a(x, 0) ,

are satisfied and that g(x, t) = ∇ · h(x, t) ,

(49)

with ht ∈ C (Ω), ∀ t ∈ [0, T ]. Then the problem (46) has a unique solution such that v ∈ C 2+α (Ω), vt ∈ C α (Ω), p ∈ C 1+α (Ω) for all t ∈ [0, T ], and α

(α)

(2+α)

sup |vt (·, t)|Ω + sup |v(·, t)|Ω t
(1+α)

+ sup |p(·, t)|Ω t
(α)

(1+α)

≤ c sup |f (·, t)|Ω + sup |g(·, t)|Ω t
t
(2+α)

+ |v0 |Ω

(1+α)

+ sup |a · n0 |Γ t
(α)

+ sup |ht (·, t)|Ω t
+[

(1+α, 1+α )

0

a]Γ ×(0,T 2)

(50)

.

Lectures on Evolution Free Boundary Problems: Classical Solutions

147

Similar theorem holds for the problem (47). Theorem 6.2 ([40]). Let Γ ∈ C 2+α , α ∈ (0, 1). For arbitrary f , g, v0 , b, B, such that v0 ∈ C 2+α (Ω), f ∈ C α (Ω), g ∈ C 1+α (Ω), b ∈ C 1+α (Γ ), 1+α B ∈ C α (Γ ) for ∀ t ∈ [0, T ], and 0 b ∈ C 1+α, 2 (Γ × (0, T )), satisfying the compatibility conditions (48) and the condition (49) with ht ∈ C α (Ω), t ∈ [0, T ], the problem (47) has a unique solution v ∈ C 2+α (Ω), p ∈ C 1+α (Ω) with vt ∈ C α (Ω), ∀ t ∈ [0, T ], and the solution satisfies the inequality (α)

(2+α)

sup |vt (·, t)|Ω + sup |v(·, t)|Ω t
(1+α)

+ sup |p(·, t)|Ω t
(1+α)

≤ c sup |f (·, t)|Ω + sup |g(·, t)|Ω t
(2+α)

+ |v0 |Ω

t
(1+α)

+ sup |b · n|Γ t
(α)

+ sup |ht (·, t)|Ω

(51)

t
+[

(1+α, 1+α )

0

b]Γ ×(0,T 2)

(α)

+ sup |B|Γ

.

t
We give here the idea of the proof of the estimates (50) and (51). It is based on the analysis of the corresponding model problems in the half-space R3+ = {x3 > 0}, namely, vt − ν ∇2 v + ∇p = f (x, t), ∇ · v = g(x, t) , x ∈ R3+ , t > 0 v|t=0 = v0 (x) ,

 ∂vj ∂v3  ν + = aj (x , t) , x = (x1 , x2 ) ∈ R2 , j = 1, 2 ,  ∂x3 ∂xj  x =0  3  ∂v3  − p + 2ν = a3 (x , t)  ∂x3 

(52)

x3 =0

and vt − ν ∇2 v + ∇p = f (x, t), ∇ · v = g(x, t) , x ∈ R3+ , t > 0 , v|t=0 = v0 (x) ,

 ∂vj ∂vj  ν + = bj (x , t) , x ∈ R2 , j = 1, 2 ,  ∂x3 ∂x3  x3 =0 
t
t  ∂v3  2  − p+2 ν +σ ∇ v3 (x , τ ) dτ  = b3 (x , t)+ B(x , τ ) dτ ,  ∂x3 0 0

(53)

x3 =0

∂ , ∂ ). We assume at first that f = 0, g = 0, v0 = 0 and that where ∇ = ( ∂x 1 ∂x2 a, b, B are given for t > 0, decay at infinity sufficiently rapidly and satisfy the compatibility conditions

148

Vsevolod A. Solonnikov

a1 (x , 0) = a2 (x , 0) = 0 ,

b1 (x , 0) = b2 (x , 0) = 0 .

(54)

Then we can reduce these problems to the boundary value problems for the system of ordinary differential equations on the half-line x3 > 0 with the aid of the Fourier–Laplace transformation
∞
−ix
·ξ
 u )(ξ, s, x3 ) = e dx e−st u(x, t) dt . R2

0

Making this transformation we obtain −ν

) d2 v )p = 0 , ) + ∇) + ν r2 v dx23

d) v3 ) ·v ) = i ξ1 v)1 + i ξ2 v)2 + = 0, ∇ dx3

x3 > 0 ,

) (ξ, s, x3 ), p)(ξ, s, x3 ) → 0 (x3 → +∞) , v

  d) vj  + i ξj v)3  =) aj , j = 1, 2 , ν  dx3 x3 =0  d) v3  − p) + 2 ν =) a3 ,  dx3 

(55)

x3 =0

) = (i ξ1 , i ξ2 , d ), r = where ∇ dx3 −ν

 s ν −1 + |ξ|2 , arg r ∈ (− π2 , π2 ] and

) d2 v ) = 0, ) + ∇p + ν r2 v dx23

) · v) = 0 , ∇

) (ξ, s, x3 ), p)(ξ, s, x3 ) → 0 (x3 → +∞) , v

  d) vj  ν + i ξj v)3  = )bj , j = 1, 2 ,  dx3

(56)

x3 =0

− p) + 2 ν

) d) v3 σ |ξ|2 B v)3 |x3 =0 = )b3 + . − dx3 s s

Both problems (55) and (56) can be solved explicitly (see [31, 38]). The solution of (55) has the form v)j = −

aj −rx3 Rjk ) ak −rx3 Sjk ) ak e−rx3 − e−|ξ|x3 (1 − δj3 ) ) e e , + + νr νrQ νQ r − |ξ|

p) = −) a3 e−|ξ|x3 +

3

k=1

3

3

k=1

k=1

Pk ) ak e−|ξ|x3 ,

Lectures on Evolution Free Boundary Problems: Classical Solutions

149

where Pj (s, ξ) = i ξj

2 r (r + |ξ|) , Q(s, ξ)

P3 (s, ξ) = 2 |ξ|2

j = 1, 2 ,

r − |ξ| , Q(s, ξ)

r s , Q(s, ξ) = r2 +|ξ| r2 +3 r ξ 2 −|ξ|3 = (r+|ξ|) M (s, ξ), M (s, ξ) = +4|ξ|2 ν r+|ξ| and Rjk , Sjk are elements of the matrices   2 ξ1 (3 r − |ξ|) ξ1 ξ2 (3 r − |ξ|) i ξ1 r (r − |ξ|)     2  R =  ξ1 ξ2 (3 r − |ξ|) ξ2 (3 r − |ξ|) i ξ2 r (r − |ξ|)     −i ξ1 r (r − |ξ|) −i ξ2 (r − |ξ|) −|ξ| r (r + |ξ|) 

 −2 r ξ1 ξ2 −i ξ1 (r2 + ξ 2 )     2 2 2   S =  −2 r ξ1 ξ2 −2 r ξ2 −i ξ2 (r + ξ )   r 2 2 −2 i ξ1 |ξ| −2 i ξ2 |ξ| r |ξ| (r + ξ ) −2 r ξ12

and it can be considered as a solution of the Neumann problem for the system of 3 heat equations vt − ν ∇2 v = −∇p ,

x ∈ R3+ , t > 0 ,

v|t=0 = 0 ,  ∂v  =d.  ∂x3 

(57)

x3 =0

Here, 3 ) aj i ξj R3k ) aj d)j = − i ξj v)3 |x3 =0 = − ) ak + , ν νr Q ν

j = 1, 2 ,

k=1

d)3 = −

2

i ξj v)j |x3 =0 =

j=1

2

i ξj j=1

2

3

i ξj Rjk ) aj − ) ak , νr νr Q j=1

(58)

k=1

and p(x, t) is a solution of the Dirichlet problem ∇2 p(x) = 0 ,

x3 > 0 ,

p(x) → 0 (|x| → ∞) ,

p|x3 =0 = −a3 + 2 ν d3 (x , t) .

(59)

150

Vsevolod A. Solonnikov

Moreover, the solution of the problem (56) can be regarded as a solution of the problem (55) with ak (x , t) related to bm (x , t) as follows aj (x , t) = bj (x , t) , j = 1, 2 ,

3 σ |ξ| ) ν M (s, ξ) ) a3 (ξ, s) = )b3 1 − +B P (s, ξ) P (s, ξ)

2

|ξ|2 2 |ξ|3 +σ − i ξj )bj , P (s, ξ) (r + |ξ|) P (s, ξ) j=1

(60)

P (s, ξ) = ν s M (s, ξ) + σ |ξ|3 . Equations (58) and (60) are equivalent to
t
3

∂ 1  dj (x , t) = aj (x , t)− K3m (x −y  , t−τ ) am (y  , τ ) dy  dτ, ν ∂x 2 j R 0 m=1 



2 t

∂ 2 Γ (x − y  , 0, t − τ ) aj (y  , τ ) dy  dτ d3 (x , t) = ∂x 2 j R 0 j=1

3
t


    Kjm (x − y , t − τ ) am (y , τ ) dy dτ , − 0

m=1

R2


t
2

∂ a3 (x , t) = b3 (x , t) + Vj (x − y  , t − τ ) b3 (y  , τ ) dy  dτ ∂x 2 j R 0 j=1
t
+ V (x − y  , t − τ ) B(y  , τ ) dy  dτ 



+

0 R2 2

j=1

where Γ (x, t) =

∂ ∂xj

1 (4 π ν t)3/2


t
0

−

e

R2 |x|2 4νt

W (x − y  , t − τ ) bj (y  , τ ) dy  dτ , (61) and Kjm , Vj , V , W are defined by

) im = Rim K νrQ σ i ξj |ξ| , V)j = P (s, ξ)

(62) νM V) = , P

0 = −σ W

2 |ξ|3 |ξ|2 − (r + |ξ|) P P

. (63)

The following propositions provide pointwise estimates of these kernels.

Lectures on Evolution Free Boundary Problems: Classical Solutions

151

) s) = K(ζ ) 1 , ..., ζn , s) is defined for all Proposition 6.1. Assume that K(ζ, complex-valued ζj = ξj + i ηj , j = 1, ..., n, and s, such that Re s + æ |Im s| ≥ −δ1 |ξ|2 , |η| ≤ δ2 |ξ| ,

ξ = 0 ,

(64)

where æ, δ1 , δ2 are positive constants, is analytic in the domain (64), homogeneous, i.e. ) ζ, λ2 s) = λm−q K(ζ, ) s) K(λ and ) s)| ≤ |K(ζ,

c |ξ|m , (|s| + |ξ|2 )q/2

where q and m are integral numbers, q > 0, m > −n. Then the Fourier– ) Laplace pre-image of K,

1 ix·ξ ) s) ds , K(x, t) = e dξ est K(ξ, x ∈ Rn , (2 π)n+1 i Rn Re s=a>0 satisfies the inequalities |Dtk Dxj K(x, t)| ≤

c tq/2−1−k (|x|2 + t)

m+n+|j| 2

|j| ≥ 0, k ≥ 0 ,

,

and K(x, t) = 0 for t < 0. Corollary 6.1. The kernels Kjm (x , t) (62) satisfy the inequalities |Dtk Dxj
Kjm (x, t)| ≤

c(k, j) t

1 2 +k

(|x |2 + t)

2+|j| 2

k ≥ 0, |j| ≥ 0 ,

,

(65)

and they vanish for t < 0. Proposition 6.2. The kernels Vj , V , Wj satisfy the inequalities |Dxj
Vj (x, t)| + |Dxj
V (x, t)| + |Dxj
Wj (x , t)| ≤

c(j, t) , (|x| + t)2+|j|

(66)

where c(j, t) are increasing functions of t, and Vj (x , t) = V (x , t) = Wj (x , t) = 0

for t < 0 .

Proposition 6.1 is proved exactly as Lemma 3.2 in [39] (see also [41], Proposition 2.1). Proposition 6.2 is proved in [40]. Using estimates (65)–(66), it can be shown by standard methods of the potential theory (see [39, 40] and the proof of (17) that

2 2



(1+α, 1+α (1+α, 1+α (1+α) 2 ) 2 ) , (67) [dk ]R2 ×(0,T ) ≤ c [aj ]R2 ×(0,T ) + sup[a3 (·, t)]R2 k=1

j=1

t
152

Vsevolod A. Solonnikov

and that for the function a3 (x , t) in (61)3 the inequality

3

(1+α) (1+α) (α) sup[a3 (·, t)]R2 ≤ c(T ) sup[bk (·, t)]R2 + sup[B(·, t)]R2 t
k=1

(68)

t
holds. Now, we can estimate the solutions of the problems (52) and (53). Theorem 6.3. If v0 ∈ C 2+α (R3+ ), f ∈ C α (R3+ ), g(·, t) ∈ C 1+α (R3+ ), a3 (·, t) ∈ 1+α C 1+α (R2 ) for all t ∈ [0, T ], aj ∈ C 1+α, 2 (R2 ×(0, T )), j = 1, 2, the compatibility conditions

 ∂v0j ∂v03  ν + = aj (x , 0), j = 1, 2 , (69) ∇ · v0 = g(x, 0) ,  ∂x3 ∂xj  x3 =0

hold and ht ∈ C α (R3+ ), ∀ t ∈ [0, T ] ,

g(x, t) = ∇ · h(x, t) ,

(70)

then the problem (52) has a unique solution v ∈ C 2+α (R3+ ), p ∈ C 1+α (R3+ ), such that vt ∈ C α (R3+ ), ∀ t ∈ [0, T ], and this solution satisfies the inequality (α)

(2+α)

(1+α)

supt
+ supt
+

(1+α)

≤ c supt
+

(1+α)

+ supt
+

(α)

+ supt
"2

(1+α, 1+α 2 ) j=1 [aj ]R2 ×(0,T )

+

,

(71)

with the constant independent of T . Theorem 6.4. If v0 , f , g, bk satisfy the hypotheses of Theorem 6.3 and B(·, t) ∈ C α (R2 ), ∀ t ∈ [0, T ], then the problem (53) has a unique solution v(·, t) ∈ C 2+α (R3+ ), p(·, t) ∈ C 1+α (R3+ ) with vt (·, t) ∈ C α (R3+ ) for all t ∈ [0, T ], and this solution satisfies the inequality (α)

(2+α)

supt
(1+α)

+supt
+

(1+α)

≤ c(T ) supt
(1+α)

+supt
+

+

"2

(α)

+supt
(1+α, 1+α (α) 2 ) j=1 [bj ]R2 ×(0,T ) +supt
+

. (72)

Lectures on Evolution Free Boundary Problems: Classical Solutions

153

We restrict ourselves with the proof of the estimates (71), (72). We assume at first that f = 0, g = 0 and consider the problem (57). Well known estimates for the parabolic Neumann problem and for harmonic functions imply   (1+α, 1+α ) (α) (2+α) (α) , ≤ c sup[∇p(·, t)]R3 + [d]R2 ×(0,T2) sup[vt (·, t)]R3 + sup[v(·, t)]R3 +

t
t
(α)

+

(1+α)

sup[∇p(·, t)]R3 ≤ c sup[a3 (·, t)]R2 +

t
+

t


t
(1+α)

 .

+ sup[d3 (·, t)]R2 t
Now, we take account of (67) and arrive at the estimate (71) with f = 0, g = 0, h = 0, from which, due to (68), inequality (72) follows (also with f = 0, g = 0, h = 0). The general case is treated by the construction of auxiliary functions satisfying non-homogeneous Stokes system. Let f ∗ and v0∗ be extensions of f (x, t) and v0 (x) into the whole space R3 such that [f ∗ (·, t)]R3 ≤ c [f (·, t)]R3+ , (α)

[v0∗ ]R3

(2+α)

(2+α)

≤ c [v0 ]R3

.

They can be defined, for instance, by the formulas f ∗ (x, t) = f (x , −x3 , t) , x3 < 0 ,   3

x3 v0∗ (x) = , λk v0 x , − k k=1

where (λ1 , λ2 , λ3 ) is a solution of the system s  3

1 λk − = 1 , s = 0, 1, 2 . k k=1

Consider the functions
t

dτ Γ (x − y, t − τ ) f ∗ (y, t) dy + Γ (x − y, t) v0∗ (y) dy , v(1) (x, t) = R3

0

R3

p(1) (x, t) = 0 , v(2) (x, t) = ∇


 R3



E(x − y) − E(x − y ∗ )

p(2) (x, t) = ν(g−∇·v(1) )+




 g − ∇ · v(1) (y, t) dy ,

% &  (1) ∇y E(x−y)−E(x−y ∗ ) h(y, t)−vt (y, t) dy,

R3+

where y = (y  , −y3 ), and E(x) = − 4 π1|x| is a fundamental solution of the Laplace equation. It is easily seen that ∂ (1) (v + v(2) ) − ν ∇2 (v(1) + v(2) ) + ∇(p(1) + p(2) ) = f , ∂t ∇ · (v(1) + v(2) ) = g ,

v(1) + v(2) |t=0 = v0 (x) ,

154

Vsevolod A. Solonnikov

so the functions u(x, t) = v − v(1) − v(2) ,

q = p − p(1) − p(2)

satisfy (52) or (53) with f = 0, g = 0. Hence, these functions can be estimated by inequalities (71) or (72) with f = 0, g = 0, obtained above. Using this inequalities and classical estimates of Newtonian potential we easily prove (71), (72) in the general case. We have considered model problems (52), (53). As for the estimates (50), (51) they are obtained by a standard Schauder’s method and the solvability of the problems (46), (47) can be proved by the construction of the regularizer (see [31, 32, 39, 40]). In addition to (71), (72), the solutions of the problems (46), (47) (with g = 0) satisfy coercive estimates (see [24, 31]), namely, (2+α,1+α/2)

(α,α/2)

(γ,1+α)

+ |∇p|Ω×(0,T ) + pΩ×(0,T ) |v|Ω×(0,T )   (1+α, 1+α ) (α,α/2) (2+α) (γ,1+α) (73) ≤ c |f |Ω×(0,T ) + |v0 |Ω + | 0 a|Γ ×(0,T 2) + a · nΓ ×(0,T ) and (2+α,1+α/2)

(α,α/2)

(γ,1+α)

+ |∇p|Ω×(0,T ) + pΩ×(0,T ) |v|Ω×(0,T )   (1+α, 1+α ) (α,α/2) (2+α) (γ,1+α) (α,α/2) ≤ c |f |Ω×(0,T ) +|v0 |Ω +| 0 b|Γ ×(0,T2 ) +b · nΓ ×(0,T ) +|B|Γ ×(0,T ) , (74) where γ ∈ (0, 1), (γ,1+α)

uΩ×(0,T ) =

sup |t − t |−

t,τ ∈(0,T )

1+α−γ 2

|u(·, t) − u(·, t )|Ω

(γ)

(75)

(γ,1+α)

and uΓ ×(0,T ) is defined in a similar way. These estimates hold under the conditions ∇ · f = 0 and B|t=0 = 0.

7 Lagrangean coordinates and local existence theorems In the problems of fluid mechanics, there exists a natural mapping of the domain Ωt with a free boundary onto a given domain Ω0 ≡ Ω. This mapping is given by a relationship between the Eulerian and the Lagrangean coordinates, i.e.,
t

u(ξ, τ ) dτ ≡ X(ξ, t) ,

x = ξ + 0

where

  u(ξ, t) = v X(ξ, t), t ,

ξ∈Ω ,

Lectures on Evolution Free Boundary Problems: Classical Solutions

ut − ν ∇u2 u + ∇u q = 0 , ∇u · u = 0, ξ ∈ Ω, t > 0 , u|t=0 = v0 (ξ) , Tu (u, q) n − σ ∆(t) X(ξ, t)|Γ = 0 ,

155

(76) (77)

where q(ξ, t) = p(X(ξ, t), t), ∇u = J −T ∇ is a transformed gradient, J = ∂xk ( ∂x )k,m=1,2,3 , Tu = −q I + ν Su (u) and Su = (∇u u) + (∇u u)T are transm formed stress and rate-of-strain tensor, respectively, ∆(t) is the Laplace– Beltrami operator on the surface Γt = X(Γ, t) .

(78)

We have used the well-known formula H n = ∆(t) X. By virtue of the equation ∇ · v = 0, the matrix J −T coincides with the matrix of cofactors A = (Akm )k,m=1,2,3 of the elements
t ∂uk (ξ, τ ) ∂xk akm (ξ, t) = = δkm + dτ (79) ∂ξm ∂ξm 0 "3 of the matrix J, hence, ∇u = A∇. Moreover, since k=1 ∂ξ∂k Akm (ξ, t) = 0, we have ∇u · u = ∇ · A u . (80) We observe finally that the normal n to Γt at the point X(ξ, t) and the normal n0 to Γ at the point ξ are related to each other by n=

A n0 . |A n0 |

(81)

It is clear that theorems 5.1 and 5.2 reduce to the proof of the solvability of the problem (76), (77) with σ = 0 or σ > 0. For subsequent computations it is convenient to separable tangential and normal components in the boundary condition (77). Let 0 f = f − n0 (n0 · f ) and f = f − n (n · f ) be projections of the vector field f (ξ), ξ ∈ Γ , onto tangent planes to Γ and Γt , respectively. If n · n0 > 0 (which is certainly the case for small t), then (77) is equivalent to two equations 0

Su (u) n = 0 ,

n0 · Tu (u, q) n − σ n0 · ∆(t) X = 0 ,

ξ∈Γ .

Making use of the identities


t

n0 · ∆(t) X = n0 · ∆(t) ξ + n0 · ∆(t) u(ξ, τ ) dτ 0
t
t ˙ ) ξ dτ u(ξ, τ ) dτ + H0 (ξ) + n0 · ∆(τ = n0 · ∆(0) 0 0


t
t   ˙ ) ∆(τ u(ξ, τ  )dτ  + ∆(t)−∆(0) u(ξ, τ ) dτ , +n0 · 0

0

156

Vsevolod A. Solonnikov

˙ where H0 (ξ) = n0 (ξ) · ∆(0) ξ is the doubled mean curvature of Γ0 and ∆(t) = d ∆(t) we write the problem (76), (77) in the form dt ut − ν ∇2 u + ∇q = F1 (u, q) , ∇ · u = F2 (u) , u|t=0 = v0 (ξ) , 0

(82)

S(u) n0 |Γ = F3 (u) ,




n0 · T (u, q) n0 − σ n0 · ∆(0)

t

u(ξ, τ ) dτ |Γ




0

= σ H0 (ξ) + F4 (u, q) + σ F5 (u) + σ

t

F6 (u) dτ , 0

where F1 (u, q) = ν (∇u2 − ∇2 ) u + (∇ − ∇u ) q , F2 (u) = (∇ − ∇u ) · u = ∇ · h(u) , h(u) = (I − AT ) u ,    F3 (u) = 0 0 S(u) n0 − Su (u) n  , Γ    F4 (u, q) = q n0 · (n − n0 ) + ν n0 · S(u) n0 − Su (u) n  , Γ
t ˙ ) ξ dτ ∆(τ F5 (u) = n0 · 0


t   ˙ u(ξ, τ ) dτ + ∆(t) − ∆(0) u(ξ, t) . F6 (u) = n0 · ∆(t)

(83)

0

We shall need the following simple estimates of the elements Aik (ξ, t) of the matrix A. Proposition 7.1. Let u(ξ, t) and u (ξ, t) be two vector fields such that
0

t


(2+α) |u(·, τ )|Ω

dτ ≤ µ , 0

t

|u (·, τ )|Ω

(2+α)

dτ ≤ µ

(84)

and let Aik and Aik be corresponding cofactors of aik . There hold the estimates
t  (1+α)   (2+α) ≤ c (1 + µ) |u − u |Ω dτ , supAik (·, τ )−Aik (·, τ ) τ
Ω

0

% &( 1+2α ) 1−α sup Aik (ξ, ·)−Aik (ξ, ·) ≤ c(1 + µ) t 2 Ω

(0,τ )

    sup ∇u(ξ, τ )−∇u (ξ, τ ) Ω×(0,t)

(85)

Lectures on Evolution Free Boundary Problems: Classical Solutions

157

and, in particular,


t

sup |Aik (·, τ ) − δik | ≤ c (1 + µ) t<τ

1−α ( 1+α 2 ) ≤ c(1+µ)t 2 sup[Aik (ξ, ·)](0,t) Ω

(2+α)

|u|Ω

0

dτ , (86)

sup sup |∇u(ξ, τ )|. Ω τ
The following proposition is a consequence of (81), (85), (86). Proposition 7.2. If u, u satisfy (84) and µ is so small that |An0 |, |A n0 | ≥ An0 A
n0 − |A c, then the difference n − n = |An
n | satisfies the inequalities 0| 0 sup |n − τ
(1+α) n |Γ ( 1+α )

2 sup[n − n ](0,t)




t

≤ c 0

≤ ct

|u − u |Ω

(2+α)

    sup ∇u(ξ, τ ) − ∇u (ξ, τ ) .

1−α 2

Γ

dτ , (87)

Ω×(0,t)

In particular,
sup |n − τ
(1+α) n0 |Γ ( 1+α )

2 sup[n − n0 ](0,t)

t

≤ c 0

≤ ct

Γ

(2+α)

|u|Ω

dτ , (88)

1−α 2

sup |∇u(ξ, τ )| . Ω×(0,t)

Similar estimates can be obtained for the coefficients of the Laplace– Beltrami operator ∆(t). We recall that in the local coordinates η1 , η2 the Laplace–Beltrami operator is defined by 2 2 2



∂2f ∂f 1 ∂ βγ √ ∂f ∆(t) f = √ g g = g βγ + hγ , g ∂ηβ ∂ηγ ∂ηβ ∂ηγ ∂ηγ γ=1 γ,β=1

β,γ=1

where g = det(gβγ )β,γ=1,2 , gβγ =

∂X ∂X · , ∂ηβ ∂ηγ

g βγ are the elements of the inverse matrix (gγβ )−1 , and 2 1 ∂ √ hγ = √ (g βγ g) . g ∂ηβ β=1

We cover Γ by the sets Γk = {ξ ∈ Γ : |ξ − ξk | < d} where d is sufficiently small number and ξk ∈ Γ (k = 1, ..., M ), take as local coordinates Cartesian

158

Vsevolod A. Solonnikov (k)

(k)

coordinates η1 = η1 , η2 = η2 on the tangential planes to Γ at the points (k) ξk , and draw the coordinate axes η3 parallel to n0 (ξk ). Let (k)

(k)

(k)

= φ(k) (η1 , η2 ) ,

η3

(k)

(k)

|η1 |2 + |η2 |2 ≤ d2 (k)

be equation of Γ in the neighbourhood of ξk . In the coordinates system η1 , (k) (k) η2 , η3


t
t
t X(ξ, t) = η1 + ω1 (η, τ ) dτ, η2 + ω2 (η, τ ) dτ, φ + ω3 (η, τ ) dτ
= ξ+

0

0

0

t

ω(η, τ ) dτ , 0

where ξ = (η1 , η2 , φ(η1 , η2 )) and ω = u(η1 , η2 , φ, t) (we omit the superscript (k)). Hence,


t

∂ω ∂ξ dτ + · ∂ηβ 0 ∂ηβ
t
t ∂ω ∂ω dτ · dτ , + 0 ∂ηβ 0 ∂ηγ

(0)

gγβ = gγβ +

∂ξ · ∂ηγ

where (0)

gγβ =


0

t

∂ω dτ ∂ηγ

(89)

∂ξ ∂ξ · , ∂ηγ ∂ηβ

and g˙ γβ =

∂ξ ∂ω ∂ξ ∂ω ∂ω · + · + · ∂ηγ ∂ηβ ∂ηβ ∂ηγ ∂ηγ


0

t

∂ω ∂ω dτ + · ∂ηβ ∂ηβ


0

t

∂ω dτ . ∂ηγ

The following proposition is evident. Proposition 7.3. Let u and u be two vector fields satisfying (84) with µ so small that g ≥ c > 0, g ≥ c > 0 ,   where g = det(gγβ ), g  = det(gγβ ) and gγβ , gγβ are functions (89) corre sponding to u and u , respectively. For the coefficients of the operators ∆(t) and ∆ (t) the following estimates hold

 (1+α)  (α)
t  γβ   (2+α) γβ    supg (·, τ )−(g )  +suphγ (·, τ )−hγ (·, τ ) ≤ c |u−u |Γk dτ , τ
Γk

τ
Γk

0

 (1+α)  (α)     (2+α) +suph˙ γ (·, τ )− h˙ γ (·, τ ) ≤ c sup |u−u |Γk , supg˙ γβ (·, τ )−(g˙ γβ )  τ
Γk

τ
Γk

τ
(90)

Lectures on Evolution Free Boundary Problems: Classical Solutions

159

in particular,
t  (1+α)  (α)     (2+α) + suphγ (·, τ ) − h0γ  ≤ c |u|Γk dτ , supg γβ (·, τ ) − g0γβ  Γk

τ
Γu

τ
0

(2+α) sup |g˙ γβ (·, τ )| + sup |h˙ γ (·, τ )| ≤ c sup |u|Γk . τ
τ
τ
(91) As a consequence of inequalities (85)–(88), (90), (91), we obtain the estimates of the expressions (83). Let (u, q) and (u , q  ) satisfy the conditions (α)

(2+α)

Nt (u, q) = sup |ut (·, t)|Ω + sup |u(·, t)|Ω τ
τ
(1+α)

+ sup |q(·, t)|Ω τ
≤ µ,

Nt (u , q) ≤ µ , u|t=0 = u |t=0 . Then     (α) (α) (α)        supF1 (u, q)−F1 (u , q  ) ≤ ν sup(∇u2 −∇u2
) u +sup∇u2
(u−u ) Ω Ω Ω τ
τ
τ
Ω

≤ c t µ Nt (u − u , q − q  ) ,    (1+α) (α) ∂       h(u) − h(u )  + sup supF3 (u) − F2 (u ) Ω τ
  (1+α) (1+α)     ≤ sup(∇u − ∇u
) · u + sup∇u
· (u − u ) τ
Ω

Ω

τ
  (α)

(α)   ∂AT (u) ∂AT (u )  ∂AT (u )       − u + sup (u − u ) + sup   ∂t ∂t ∂t τ
Ω

  (α)  (α)     + sup AT (u) − AT (u ) ut  + sup(I − AT (u )) (ut − ut ) τ
Ω

τ
Ω

≤ c (t + t1/2 ) µ Nt (u − u ) (92)

160

Vsevolod A. Solonnikov

(we have used the inequality    ∂AT (u) ∂AT (u ) (α)  (α)     − ≤ c sup∇(u − u ) ≤ c t1/2 Nt (u − u ) , sup   ∂t ∂t Ω τ
Nt (u − u ) ≡ Nt (u − u , 0)

).

Further we have   (1+α) (1+α)     + supF4 (u, q) − F4 (u , q  ) supF3 (u) − F3 (u ) Γ

τ
Γ

τ




≤ c t µ Nt (u − u , q − q ) , % &( 1+α 2 ) sup F3 (u)−F3 (u ) (0,t)

Γ

  1−α ( 1+α 2 ) ≤ c t µ sup[∇(u−u )](0,t) + t 2 sup |∇u| sup |∇(u−u )| Γ ×(0,t)

Γ

(93)

Γ ×(0,t)



≤ c t µ Nt (u − u ) ,  (α)   ≤ c t µ Nt (u − u ) . supF6 (u) − F6 (u ) Γ

τ
∂ξ ˙ ) ξ do not give any = 0, the lower order terms of ∆(τ Finally, since n0 · ∂η ν contribution into Fs (u), and we have


t  (1+α)   (2+α) supF5 (u) − F5 (u ) ≤ c |u − u |Ω dτ ≤ c t Nt (u − u ) . Γ

τ
0

In particular, if u = 0, q  = 0, then (α)

(1+α)

sup |F1 (u, q)|Ω + sup |F2 (u)|Ω τ
τ
(1+α)

+ sup |F4 (u)|Γ τ
and, if u|t=0 = v0 (ξ), then

τ
(α)

+ σ sup |F6 (u)|Γ τ

(1+α) sup |F5 (u)|Γ τ
(1+α)

+ sup |F3 (u)|Γ

≤ c

≤ c t Nt2 (u, q) ,

t

Nt (u) dτ ≤ c t Nt (u) , 0

(94)

(95)

Lectures on Evolution Free Boundary Problems: Classical Solutions

161

 (α)   ∂   supτ


≤ c t ≤ ct

1−α 2

1−α 2

sup |∇u(ξ, t)| + t N (u) 2

2

(96)



Γ ×(0,t)

sup |∇v0 (ξ)|2 + c t N 2 (u) . Γ

Now, we can give the proof of Theorems 5.1 and 5.2. For definiteness, we prove the second theorem; the same arguments (with σ = 0) provide the proof of Theorem 5.1. Proof of Theorem 5.2. We write the problem (82) in an equivalent form (u, q) = F(u, q) + (u0 , q0 ) ≡ H(u, q) ,

(97)

where (u0 , q0 ) is a solution of the problem (47) with f = 0, g = 0, B = 0, b = σ H0 n0 and F(u, q) ≡ (w, r) is a solution of the same problem with f = F1 (u, q), g = F2 (u), v0 = 0, b = F3 (u) + n0 (F4 (u, q) + σ F5 (u)), B = σ F6 (u). The solution of the equation (97) is sought in the space Cα (Ω×(0, T )) with the norm equal to NT (u, q). We show that H(u, q) is a contraction operator in a subset of a certain ball NT0 (u, q) ≤ µ , T0  1 , of this space, whose elements satisfy the initial condition u|t=0 = v0 (ξ). Indeed, by virtue of (94)–(96), for arbitrary element of the above subset we have H(u, q)Cα ≤ NT0 (w, r) + NT0 (u0 , q0 )   1−α ≤ NT0 (u0 , q0 ) + c1 sup |∇v0 (ξ)| |v0 (ξ)| + T0 2 sup |∇v0 (ξ)|2  Ω 1+α  + c2 NT0 (u, q) T0 + T0 2 sup |v0 (ξ)| + T0 N (u, q) Ω

and, if



1−α 2

µ ≥ 2 NT0 (u0 , q0 ) + c1 sup |∇v0 (ξ)| |v0 (ξ)| + T0 Ω

and

sup |∇v0 (ξ)|

2



Γ

  1+α 1 c2 T0 + T0 2 sup |v0 (ξ)| + T0 µ ≤ , 2 Ω

(98)

162

Vsevolod A. Solonnikov

then H(u, q) belongs to the same ball H(u, q)Cα (Ω) ≤ µ . Further, by virtue of (92), (93), for arbitrary (u, q), (u , q  ) in this ball such that u|t=0 = u |t=0 = v0 there holds 1 1 1 1 ≤ c3 (T0 + T0 µ) NT0 (u − u , q − q  ) . 1H(u, q) − H(u , q  )1 Cα (Ω)

So, if c3 T0 (1 + µ) < 1 , then H(u, q) is a contraction operator and equation (96) has a solution in the above ball. The uniqueness of the solution (u, q) ∈ Cα (Ω×(0, T0 )) of the problem (82) follows from the inequalities (51), (93)–(95). The theorem is proved. Remark 7.1. Theorem 5.2 holds for arbitrary domain Ω ⊂ R3 with a smooth boundary. If Ω is homeomorphic to the ball, it has a sense to introduce a 1/3 new pressure p = p − R20 where R0 = ( |Ω| , and then instead of (38) we 4π ) shall have  (1+α)  2  (α) (2+α) + sup p(·, t) − sup |v(·, t)|Ωt + sup |v(·, t)|Ωt R 0  Ωt t
The functions (u0 , q0 ) in (96) will be a solution of (47) with b = σn0 (H0 + 2 R0 ), and slightly more accurate estimates of F5 (u) will lead, instead of (98), to the inequality  1+α  1 c ecT0 T0 2 sup |v0 (ξ)| + T0 µ ≤ , 2 Ω which shows that T0 → ∞ as the norms of v0 and H0 + as the domain Ω0 approaches a ball).

2 R0

tend to zero (i.e.

8 Proof of Theorem 5.3 We start with the following auxiliary propositions. Proposition 8.1. Let G be a bounded domain in R3 satisfying the cone condition. For arbitrary vector field u ∈ W21 (G) the Korn inequality 


2 
% &     u×(x−x0 ) dx (99) u2W 1 (G) ≤ c1 S(u)2L2 (G) + c2  u dx +  2

holds.

G

G

Lectures on Evolution Free Boundary Problems: Classical Solutions

163

Korn’s inequality was proved under different assumptions on the domain G and on the vector fields u(x) by many authors (see, for instance, [16, 19, 20]). We use the variant of this inequality obtained in [35]. Proposition 8.2. Assume that a classical solution of the problem (35) with σ = 0 and with v0 (x) satisfying the conditions (36) is defined for t ∈ [0, T ], and let the Korn inequality (99) be satisfied for G = Ωt with the constant c1 independent of t. Then v(·, t)2L2 (Ωt ) ≤ e−2bt v0 2L2 (Ω) .

(100)

Proof. Multiplication of the equation (35)1 by v(x, t) and integration over Ωt gives 1 d ν v(·, t)2L2 (Ω) + S(v)2L2 (Ω) = 0 . 2 dt 2 The vector field v satisfies the condition

v · η(x) dx = v0 · η dx = 0 , Ωt

Ω

where η = a + b×x is an arbitrary vector field of rigid displacements (which can be easily verified by multiplication of (5.51 ) by η and integration). Hence,

v dx = [v×x] dx = 0 , Ωt

Ωt

and by virtue of (99), d v2L2 (Ω) + 2 b v2L2 (Ω) ≤ 0 dt with 2 b = ν c−1 . This inequality implies (100). Remark 8.1. It follows from the Proposition 8.2 that an arbitrary solution of (35). defined in an infinite time interval decays exponentially, if (99) holds with the constant independent of t. To complete the proof of Theorem 5.3, we should get estimates similar to (100) for the H¨ older norms of the solution of the problem (35). Proposition 8.3. Assume that the solution of the problem (35), with σ = 0 is defined for t ∈ (0, T ) and NT (v, p) = sup |vt (·, t)|Ωt + sup |v(·, t)|Ωt (α)

(2+α)

t
t
(1+α)

+ sup |p(·, t)|Ωt

≤ µ.

t
Then there holds the inequality sup

(α)

|vt (·, τ )|Ωτ +

τ ∈(t0 −τ0 ,t0 )

|v(·, τ )|Ωτ +

sup

τ ∈(t0 −τ0 ,t0 )

≤ c(τ0 ) where t0 ∈ (0, T ) and τ0 ∈ (0,

sup

τ ∈(t0 −τ0 ,t0 )

sup

v(·, τ )L2 (Ω) ,

τ ∈(t0 −2τ0 ,t0 )

t0 2)

(1+α)

|p(·, τ )|Ωτ

is a number depending on µ.

(101)

164

Vsevolod A. Solonnikov

Proof. Let ζλ (t) be a smooth monotone function of t equal to zero for t < t0 − 2 τ0 + λ2 , to one for t > t0 − 2 τ0 + λ, λ ∈ (0, τ0 ), and such that    dζ (t)   λ   ≤ c λ−1 .   dt  The functions w = v ζλ , r = p ζλ satisfy the equation wt − ν ∇2 w + (v · ∇) w + ∇r = v ζλ , ∇ · w = 0,

x ∈ Ωt , t > t0 − 2 τ0 ,

(102)

w|t=t0 −2τ0 = 0 , T (w, q) n|Γt = 0 .

We introduce the Lagrangean coordinates, according to the formula
t u(ξ, τ ) dτ ≡ X(ξ, t) , t > t0 − 2 τ0 , x = ξ+ t0 −2τ0

where u(ξ, t) = v(X(ξ, t), t), and write (102) in the form wt − ν ∇2 w + ∇r = u ζλ + F1 (w, q, u) , ∇ · w = F2 (w, u) , 0

ξ ∈ Ω  , t > t0 − 2 τ0 ,

S(w) n0 |Γ
= F3 (w, u) ,

n0 · T (w, r) n0 |Γ
= F4 (w, u) , w|t=t0 −2τ0 = 0 , where

F1 (w, q, u) = ν (∇u2 − ∇2 ) w − (∇u − ∇)r , F2 (w, u) = (∇ − ∇u ) · w = ∇ · (I − AT ) w ,    F3 (w, u) = 0 0 S(w) n0 − Su (w) n  , Γ    F4 (w, u) = ν n0 · S(w) n0 − n · Su (w) n  , Γ







have the same Ω = Ωt0 −2τ0 , Γ = ∂Ω , n0 is the normal to Γ , 0 and meaning as in Section 7 (we have used the same notations for the functions w and r expressed in Eulerian and Lagrangean coordinates). By virtue of inequalities (50), (86), (87) and of the fact that w|t=t0 −2τ0 = 0, there holds the estimate

Lectures on Evolution Free Boundary Problems: Classical Solutions

Nt0 (w, r) ≡

(α)

sup

t0 −2τ0 <τ
+

|wτ (·, τ )|Ω
+

(2+α)

sup

t0 −2τ0 <τ
|w(·, τ )|Ω


(1+α)

sup

t0 −2τ0 <τ
|r(·, τ )|Ω


(103)



≤ c τ0 µ Nt0 (w, r) + λ−1

165

sup

(α)

|v(·, τ )|Ω

t0 −2τ0 <τ
 .

We subject τ0 to the requirement c τ0 µ ≤

1 2

and use the interpolation inequality (α)

|v(·, τ )|Ω

(2+α)

≤ θ2 |v(·, τ )|Ω


+ c θ−α−3/2 v(·, τ )L2 (Ω) ,

valid for arbitrarily small θ. Setting θ = λ1/2 ε1/2 , we easily see that (103) implies λ +K , (104) f (λ) ≤ c1 ε f 2 where f (λ) = λα+5/2 K = c(ε)



(α)

(2+α)

sup |uτ (·, τ )|Ω
+ sup |u(·, τ )|Ω


τ ∈Iλ

sup

τ ∈Iλ

u(·, τ )2L2 (Ω
) ,

t0 −2τ0
Taking ε =

1 2 c1

(1+α)

+ sup |q(·, τ )|Ω




τ ∈Iλ

,

Iλ = (−t0 − 2τ0 + λ, t0 ) .

and making iterations we easily obtain f (λ) ≤ 2 K ,

(105)

which yields (101), if we fix λ = τ0 . The proposition is proved. It follows from (101) and (100) that (α)

(2+α)

|vt (·, t0 )|Ωt + |v(·, t0 )|Ωt 0

0

(2+α)

+ |p(·, t0 )|Ωt

0

≤ c(τ0 ) e−2t0 v0 L2 (Ω) .

If c(τ0 ) v0 L2 (Ω) ≤ µ (this inequality is satisfied, if the number ε in (39) is sufficiently small), then we obtain a uniform estimate of the norm Nt (v, p) and by a repeated application of the local existence theorem we can extend the solution into an infinite time interval t > 0. The theorem is proved. Due to the exponential decay of the solution, the transformation X is well defined for all t > 0. The free boundary Γt has a limit Γ∞ = X(Γ, ∞) ∈ C 2+α . Examples show (see [9]) that in the case β = 0 Γt can have no definite limit as t → ∞.

166

Vsevolod A. Solonnikov

9 Scheme of the proof of Theorem 5.4 Theorem 5.4 is proved according to the same scheme as Theorem 5.3: at first the energy inequality is established, and then local estimate of the type (100) is obtained for the functions w(x, t) = v(x, t) − v∞ (x) = v(x, t) +

β h(x) , I∞

β2 (x21 + x22 ) − p1 , 2 2 I∞

s(x, t) = p(x, t) − p∞ (x) = p(x, t) −

where β and p1 are the same constants as in Section 1, h(x) = (−x2 , x1 , 0) ,
I∞ = (x21 + x22 ) dx ,

 Ω∞ =

 |x| < R∞

Ω∞

x |x|

 .

It is easily seen that w, s satisfy the relations wt + (v · ∇) w − ∇ · w = 0,

β (w · ∇) h − ν ∇2 w + ∇s = 0 , I∞

x ∈ Ωt , t > 0 ,

x∈Ω , w|t=0 = v0 (x) − v∞ (x) ≡ w0 (x) ,

β 2 |h|2 T (w, s) n = σ n H + + p1 , 2 2 I∞ β h · n , x ∈ Γt . Vn = w · n − I∞

(106)

In addition, we have


2 β

It [w×x] dx = ej x3 xj dx + e3 β 1 − I∞ j=1 I∞ Ωt Ωt





2   β

5 = ej ω3 ωj R5 (ω, t) − R∞ (ω) dω 5 I∞ j=1 S1
  β 5 R5 (ω, t) − R∞ e3 (ω) (ω12 + ω22 ) dω , − 5 I∞ S1


where

|h|2 dx =

It = Ωt

1 5


(ω12 + ω22 ) R5 (ω, t) dω , S1

(107)

Lectures on Evolution Free Boundary Problems: Classical Solutions

167

moreover, if we assume that the barycenter of Ω coincides with the origin  and Ω v0 dx = 0, then

β w dx = h(x) dx = 0 . (108) I∞ Ωt Ωt We present here only main ideas of the proof of Theorem 5.4 which is more technical than the proof of Theorem 5.3. The details will be given in a forthcoming paper of the author and Prof. M. Padula. Proposition 9.1. Assume that a classical solution of the problem (106) with σ > 0 is defined for t ∈ (0, T ), the free boundary is given by equation (44), and the Korn inequality (99) for G = Ωt holds with the constant independent of t. Moreover, let the momentum β be small enough and let sup |R(ω, t) − R∞ (ω)| + sup |∇S1 (R − R∞ )| ≤ δ  1 .

(109)

w(·, t)2L2 (Ωt ) + R(·, t) − R∞ (·)2W 1 (S1 ) 2   −2bt 2 w0 L2 (Ωt ) + R(·, 0) − R∞ 2W 1 (S1 ) , ≤ ce

(110)

S1 ×(0,T )

S1 ×(0,T )

Then

2

with the constants b and c independent of T . Scheme of the proof. Multiplication of equation (106)1 by w and integration over Ωt leads to 1 d ν w(·, t)2L2 (Ωt ) + S(w)2L2 (Ωt ) 2 dt 2


2 β |h(x)|2 H+ = σ + p1 w · n dS 2 2 I∞ Γt
  = σ H[R]−H[R∞ ] R2Rt dω

(111)

S1


  √ σ β2 2 2 2 2 (ω +ω ) R (ω, t)−R (ω) R g w · n||x|=R(ω,t) dω 1 2 ∞ 2 2 I∞ S1
  d β −σ H[R]−H[R∞ ] R2 ω12 +ω22 (R−R∞ )dω ≡ I1 +I2 +I3 , I∞ S1 dϕ +

 where g = R2 + |∇S1 R|2 and ϕ is the  angle of rotation about the x3 -axis (we have used the relation h · n|Γt = x21 + x22 nϕ and the fact that R∞ is independent of ϕ). As shown in [42], I1 + I2 = −

d E1 (R−R∞ )+J1 (R−R∞ ) , dt

168

Vsevolod A. Solonnikov

where E1 (R − R∞ ) =


R |∇S1 (R − R∞ )|2 √ dω − (R − R∞ )2 dω g S1 S1
 2 R 1   + ∇ω R∞ · ∇ω (R − R∞ ) 3/2 dω 2 S1 g


R2 R + dω ∇ω R∞ · ∇ω (R − Rω ) √ − √ g R∞ g∞ S1 1 2




and

  |J1 | ≤ c (δ + β 2 ) R − R∞ W21 (S1 ) wL2 (Γt ) + R − R∞ 2W 1 (S1 ) . (112) 2

The integral I3 can also be transformed by integration by parts and estimated by c β δ R − R∞ 2W 1 (S1 ) with the constant c depending on the second 2 derivatives of R. Finally, the Korn inequality and equations (107), (108) give S(w)2L2 (Ωt ) ≥ c1 w2W 1 (Ωt ) − c2 β 2 R − R∞ 2L2 (S1 ) . 2

Hence, relation (111) implies

νc d 1 w2L2 (Ωt ) + E1 (R − R∞ ) + w2W 1 (Ωt ) + J ≤ 0 , 2 dt 2 2

(113)

with J satisfying (112). Now, following the idea of M. Padula [25], we obtain an additional estimate of R − R∞ 2W 1 (S1 ) . We introduce an auxiliary vector field V(x, t) such 2 that ∇ · V(x, t) = 0 , x ∈ Ωt , (114) r − r¯ V · n|Γt = √ ≡ f (x, t) , R g  x x , t) − R∞ ( |x| ), r¯ = 41π S1 r(ω, t) dω, and we require that V where r = R( |x| should satisfy the inequalities VW21 (Ωt ) ≤ c f W 1/2 (Γt ) ≤ c R − R∞ W21 (S1 ) , 2   Vt L2 (Ωt ) ≤ c ft L2 (S1 ) + f L2 (Γt )   ≤ c w · nL2 (Γt ) + R − R∞ W21 (S1 ) (we recall that R√Rgt = w · n − Iβ∞ h · n and h · n = satisfies the necessary condition

(115)

 x21 + x22 nϕ ). Since f

Lectures on Evolution Free Boundary Problems: Classical Solutions




169


(r − r¯) dω = 0 ,

f (x, t) ds = Γt

S1

the vector field V(x, t) with the properties (114) exists, and inequalities (115) can be satisfied. Multiplying (106)1 by V and integrating we obtain

  d w · V dx− w · Vt +(v · ∇)V dx dt Ωt Ωt
(116) ν + S(w) : S(V) dx+E2 (R−R∞ ) = 0 , 2 Ωt where E2 (R − R∞ ) = −




 H[R] − H[R∞ ] (r − r¯) dω

S1

β2 − 2 2 I∞


S1

2 (R2 (ω, t) − R∞ (ω)) (ω12 + ω22 ) (r − r¯) dω .

We add (113) and (116) multiplied by a small parameter γ, to obtain


d 1 wL2 (Ωt ) + γ w·V dx + E1 (R − R∞ ) dt 2 Ωt
νc γν 2 (117) + wW 1 (Ωt ) +J +γE2 (R−R∞ )+ S(w) : S(V) dx 2 2 2 Ωt
− γ w · (Vt +(v · ∇)V) dx ≤ 0. Ω

It may be shown that the differences E1 (R − R∞ ) −

1 E0 (R − R0 ) 2

where E0 (R − R∞ ) =

and

E2 (R − R∞ ) − E0 (R − R∞ ) ,




 |∇S1 (R − R∞ )|2 − 2 (R − R∞ )2 dω

S1

satisfy the inequality (112) and that 
2   E0 (R − R∞ ) ≥ 4 R − R∞ 2W 1 (S1 ) − c1  (R − R∞ ) dω  2 S1 
2 3

   (R − R∞ ) ωj dω  , − c2   j=1

S1

so that E0 (R−R∞ ) is positive definite for small β and δ (see [34, 37]). Hence, E1 and E2 are also positive definite and there exists such constant b > 0 that

170

Vsevolod A. Solonnikov



νc w2W 1 (Ωt ) + J + γ E2 (R − R∞ ) + 2 2



γν S(w) : S(V) dx − γ w · (Vt + (v · ∇)V) dx + 2 Ωt Ω
1 ≥ w2L2 (Ωt ) + γ w · V dx + E1 (R − R∞ ) =: E(t) 2 Ω

−1

(2 b)

≥ c1 w2L2 (Ωt ) + c2 R − R∞ 2W 1 (S1 ) , 2

if small parameters γ, δ, β are chosen in an appropriate way. Then (117) implies d E(t) + 2 b E(t) ≤ 0 dt and, as a consequence, we obtain (110). Proposition 9.2. Let the solution of the problem (31) be defined for t ∈ (0, T ) and let NT (w, s) ≤ µ . Then sup

(α)

|wτ (·, τ )|Ωτ +

t0 −τ0 <τ
sup

(2+α)

|w(·, τ )|Ωτ

t0 −τ0 <τ
+

(1+α)

(3+α)

sup |R(·, τ ) − R∞ |S1 sup |s(·, t)|Ωτ + t0 −τ0 <τ
+

t0 −τ0 <τ
t0 −2τ0 <τ
t0 −2τ0 <τ
(118) where t0 ∈ (0, T ) and τ0 <

t0 2

is a certain small number depending on µ.

Scheme of the proof. We introduce the Lagrangean coordinates which are related to the Eulerian coordinates x ∈ Ωt as follows dx(ξ, t) β = v(x(ξ, t), t) = − h(x(ξ, t)) + w(x(ξ, t), t) , dt I∞ t ≥ t0 = t0 − 2 τ0 ,

x(ξ, t0 ) = ξ .

Integrating this equation we arrive at x(ξ, t) = Z(t) y(ξ, t) ≡ X(ξ, t) , where

Lectures on Evolution Free Boundary Problems: Classical Solutions

171

 cos β  (t − t0 ) − sin β  (t − t0 ) 0 Z(t) =  sin β  (t − t0 ) cos β  (t − t0 ) 0 , 0 0 1 β , β = − I∞
t y(ξ, t) = ξ + Z −1 (τ ) u(ξ, τ ) dτ , u(ξ, t) = w(X(ξ, t), t) . 

t
0

In these coordinates the problem (106) has the form ut + β  (u · ∇u )h(X) − ν ∇u2 u + ∇u q = 0 , ∇u · u = 0 ,

ξ ∈ Ωt0 ≡ Ω  , t > t0 ,

u|t=t
0 = w(ξ, t0 ) , Tu (u, q) n = σ ∆(t) X(ξ, τ ) + p∞ (X) n ,

ξ ∈ Γ  ≡ ∂Ω  ,

where q(ξ, t) = s(X(ξ, t), t) ,

∇u = A ∇ ,

A=ZB ,

B is the matrix of cofactors of bkj (ξ, t) =

3
t

∂yk ∂um (ξ, τ ) = δkj + Z km (τ ) dτ ,
∂ξj ∂ξj m=1 t0

Z km are elements of the inverse matrix Z −1 . Following the proof of Proposition 8.3, we introduce the functions uλ (ξ, t) = ζλ (t) u(ξ, t), qλ (ξ, t) = ζλ (t) q(ξ, t). They satisfy the relations uλt + β  (uλ · ∇u ) h(X) − ν ∇u2 uλ + ∇u qx = u ζλ , ∇u · uλ = 0 , uλ |t=t
0 = 0 , 0

Su (uλ ) n|Γ
= 0 ,




n0 · Tu (uλ , qλ ) n − σ n0 ·

t

∆(τ ) uλ (ξ, τ ) dτ |Γ
= b(ξ, t) + 0




t

B(ξ, τ ) dτ ,

t
0

b = (I − Z(t)) n0 · Tu (uλ , qx ) n , ˙ ξ B(ξ, t) = ζλ (t) Zn0 · Tu (u, q) n+σ ζλ n0 · ∆(t)
t d ˙ Z −1 (τ ) u(ξ, τ )dτ +σ ζλ (t) (p∞ (X) n0 · n) . +σ ζλn0 · ∆(t) dt t
0

172

Vsevolod A. Solonnikov

The last equation is obtained from  
t   ζλ (τ ) n0 · Tu n − σ ∆(τ ) X dτ ζλ n0 · Tu n − σ ∆(t) X − t
0




= σ ζλ p∞ (X) n0 · n − σ

t

t
0

ζλ (τ ) p∞ (X) n0 · n dτ ,

by integration by parts in some terms. Making use of the inequality (51) and of the smallness of τ0 , we obtain the following estimate of (uλ , qλ ) (α)

(2+α)

sup |uλτ (·, τ )|Ω
+ sup |uλ (·, τ )|Ω

t
0 <τ
−1

t
0 <τ


(1+α)

+ sup |qλ (·, τ )|Ω
t
0 <τ


(1+α) (1+α) (α) sup |u(·, τ )|Ω
+ sup |u(·, τ )|Ω
+ sup |q(·, τ )|Γ

+λ <τ
≤c λ

(119) (cf. the proof of (61) in [42]). We estimate the norm of q using the boundary condition in (106), i.e.,


2 2 + x β2 2 x ∂w 2 − σ H − H∞ + (R − R∞ (120) ) 1 2 2 , s = 2ν n · ∂n 2 |x| and interpolation inequalities, as follows   (α) (1+α) (2+α) |s(·, t)|Γt ≤ c |w|Γt + |R − R∞ |S1   (2+α) (3+α) ≤ θ1 |w|Γt + |R − R∞ |S1   − 5 −α + c θ1 2 R − R∞ W21 (S1 ) + wL2 (Ωt ) . Here θ1 > 0 is arbitrarily small. Since linearized equation (120) (with respect to R − R∞ ) is elliptic, there holds the estimate   (3+α) (1+α) (2+α) (121) ≤ c |s|Γt + |w|Ωt + R − R∞ L2 (S1 ) . |R − R∞ |S1 Taking θ1 = λ ε , ε  1, we easily obtain from (119) inequality similar to (104)   λ  +K , f (λ) = c ε f 2 where α+ 72

f (λ) = λ

K = c(ε )





 sup

τ ∈Iλ

sup

(α) |uτ (·, τ )|Ω


+ sup

τ ∈Iλ

(2+α) |u(·, τ )|Ω


u(·, t)L2 (Ω) +

t0 −2τ0 <τ
sup

+ sup

τ ∈Iλ

(1+α) |q(·, τ )|Ω


R − R∞ W21 (S1 )

t0 −2τ <τ
 .

,

Lectures on Evolution Free Boundary Problems: Classical Solutions

173

It implies (105) and, by virtue of (121), also the desired estimate (118). In addition, we have an estimate in a small interval (0, 2τ0 )  (3+α)     (2+α) N2τ0 (w, s) ≤ c |w0 |Ω . + R(·, 0) − R∞ (·) S1

This inequality, together with (110) and (118), yields a uniform estimate for NT (w, s) which makes it possible to extend the solution of the problem (106) to an infinite time interval t > 0 and to complete the proof of Theorem 5.4.

References [1] [2] [3]

[4] [5] [6]

[7]

[8] [9] [10]

[11]

[12] [13] [14]

G. Allain, Small time existence for the Navier–Stokes equations with a free surface, Appl. Math. Optim., 16 (1987), 37–50. B.V. Bazalii, The Stefan problem, Dokl. Akad. Nauk Ukrain. SSR Ser. A, 11 (1986), 3–7. B.V. Bazalii and S.P. Degtiarev, On the classical solvability of the multidimensional Stefan problem for the convective notion of viscous incompressible fluid, Mat. Sb., 132 (1987), 3–19. J.T. Beale, Large-time regularity of viscous surface waves, Arch. Rational Mech. Anal., 84 (1984), 307–352. J.T. Beale and T. Nishida, Large-time behaviour of viscous surface waves, Lecture Notes in Num. Appl. Anal., 8 (1985), 1–14. G.I. Bizhanova, Solution in a weighted H¨ older function space of the multidimensional two-phase Stefan and Florin problems for second order parabolic equation in a bounded domain, Algebra Anal., 7 (2) (1995), 46–76. G.I. Bizhanova and V.A. Solonnikov, Solvability of an initial-boundary value problem for second order parubolic equation with time derivative in the boundary condition in a weighted H¨ older function space, Algebra Anal., 5 (1) (1993), 109–142. G.I. Bizhanova and V.A. Solonnikov, On free boundary problems for the second order parabolic equations, Algebra Anal., 12 (6) (2000), 98–139. V.O. Bytev, Non-steady motion of a ring of a viscous incompressible liquid with free boundaries, J. Appl. Math. Techn. Phys., 3 (1970), 82–88. X. Chen and F. Reitich, Local existence and uniqueness of solutions of the Stefan problem with surface tension and kinetic undercooling, J. Math. Anal. Appl., 164 (1992), 350–362. X. Chen, J. Hong and F. Yi, Existence, uniqueness and regularity of classical solutions of the Mullin–Sekerka problem, , Comm. Part. Diff. Equat., 21 (1996), 1705–1727. I.V. Denisova, Problem of the motion of two viscous incompressible fluids separated by a closed free interface, Acta Appl. Math., 37 (1994), 31–40. J. Escher and G. Simonett, Classical solutions of multi-dimensional Hele–Shaw models, SIAM J. Math. Anal., 28 (1997), 1028–1047. J. Escher and G. Simonett, Moving surfaces and abstract parubolic evolution equations, in ”Topics in Nonlinear Analysis” – The Herbert Anmann anniversary volume, Progress in nonlinear differential equations and their applications, 35 (1999), 184–210.

174

Vsevolod A. Solonnikov

[15] A. Friedman, Variational Principles and Free Boundary Problems, Wiley, New York, 1982. [16] K.O. Friedrichs, On the boundary value problems of the theory of elasticity and Korn’s inequality, Ann. of Math., 48 (1947), 441–471. [17] E.I. Hanzawa, Classical solutions of the Stefan problem, Tohoku Math. J., 33 (1981), 297–335. [18] D. Kinderlehrer and L. Nirenberg, The smoothness of the free boundary in the one phase Stefan problem, Comm. Pure Appl. Math., 31 (1978), 257–282. [19] V.A. Kondratiev and O.A. Oleinik, Hardy’s and Korn’s type inequalities and their applications, Rend. Mat. Appl., 7 (1990), 641–666. ¨ [20] A. Korn, Uber einige Ungleichungen, welche in der Theorie der elastischen und elektrischen Schwingungen eine Rolle spielen, Bull. Intern. Cracovie Akad. (Cl. Sci. Math. Nat.), (1909), 705–724. [21] S. Luckhaus, Solutions for the two-phase Stefan problem with Gibbs–Thomson law for the melting temperature, Euro. J. Appl. Math., 1 (1990), 101–111. [22] A.M. Meirmanov, On a classical solution of multidimensional Stefan problem for quasilinear parabolic equations, Matem. Sb., 112 (1980), 170–192. [23] A.M. Meirmanov, The Stefan Problem, De Gruyter, Berlin, 1992 (Russian edition: Novosibirsk, 1986). [24] I.Sh. Mogilevskii and V.A. Solonnikov, On the solvability of an evolution free boundary problem for the Navier–Stokes equations in the H¨ older spaces of functions, in “Math. Problems Related to Navier–Stokes Equations”, Series on Advances in Math. for Appl. Sciences, 11 (1992), 105–181. [25] M. Padula, On the exponential stability of the rest state of a viscous compressible fluid, J. Math. Fluid Mech., 1 (1999), 62–77. [26] E.V. Radkevich, Solvability of general nonstationary problems with free boundary, in “Some Applications of Functional Analysis to the Problems of Mathematical Physics”, AN SSSR, Sibirsk. Otdel., Inst. Mat., Novosibirsk (1986), 85–111. [27] E.V. Radkevich, On conditions for the existence of a classical solution of the modified Stefan problem (the Gibbs–Thomson law), Matem. Sb., 183 (2) (1993), 77–101. [28] J.F. Rodrigues, The variational inequality approach to the one-phase problem, Acta Appl. Math., 8 (1987), 1–35. [29] J.F. Rodrigues, V.A. Solonnikov and F. Yi, On a parabolic system with time derivative in the boundary conditions and related free boundary problems, Math. Ann., 315 (1999), 61–95. [30] V.A. Solonnikov, Estimates of solutions of nonstationary linearized system of the Navier–Stokes equations, Trudy Math. Inst. Steklov, 64 (1964), 213–317. [31] V.A. Solonnikov, Estimates of solutions of initial-boundary value problem for a linear nonstationary system of the Navier–Stokes equations, Zapiski Nauchn. Semin. LOMI, 59 (1976), 178–254. [32] V.A. Solonnikov, On the solvability of the second initial-boundary value problem for the linear time-dependent system of the Navier–Stokes equations, Zapiski Nauchn. Semin. LOMI, 69 (1977), 200–218. [33] V.A. Solonikov, Solvability of the problem on a motion of a viscous incompressible liquid bounded by a free surface, Izvestia Acad. Nauk USSR Ser. Matem., 41 (6) (1977), 1388–1427. [34] V.A. Solonnikov, On nonstationary motion of a finite liquid mass bounded by a free surface, Zapiski Nauchn. Semin. LOMI, 152 (1986), 137–157.

Lectures on Evolution Free Boundary Problems: Classical Solutions

175

[35] V.A. Solonnikov, On the transient motion of an isolated mass of a viscous incompressible fluid, Izvestia Acad. Nauk USSR Ser. Matem., 51 (5) (1987), 1065–1087. [36] V.A. Solonnikov, On the evolution of an isolated volume of viscous incompressible capillary fluid for large values of time, Vestnik Leningrad Univ. Ser. I, 3 (1987), 49–55. [37] V.A. Solonnikov, On nonstationary motion of a finite isolated mass of selfgravitating fluid, Algebra Anal., 1 (1) (1989), 207–249. [38] V.A. Solonnikov, On initial-boundary value problem for the Stokes system arising in the study of a free boundary problem, Trudy Mat. Inst. Steklov, 188 (1990), 150–188. [39] V.A. Solonnikov, Estimates of solutions of the second initial-boundary value problem for the Stokes system in the space of functions with H¨ older continuous derivatives with respect to spatial variables, Zapiski Nauchn. Semin. POMI, 259 (1999), 254–279. [40] V.A. Solonnikov, On the justification of quasistationary approximation in the problem of motion of a viscous capillary drop, Interfaces and Free Boundaries, 1 (2) (1999), 125–174. [41] V.A. Solonnikov, Initial-boundary value problem for generalized Stokes systen in the half-space, Zap. Nauchn. Semin. POMI, 271 (2000), 224–275. [42] V.A. Solonnikov, On the solvability of a quasistationary problem for the Navier–Stokes equations, Adv. Math. Sci. Appl., 11 (1) (2001). [43] V.A. Solonnikov and E.V. Frolova, Lp -theory for the Stefan problem, Zap. Nauchn. Semin. POMI, 243 (1997), 299–323. [44] V.A. Solonnikov and A. Tani, Evolution free boundary problem for equation of motion of viscous compressible barotropic liquid, Proceedings Oberwolfach, 1991, Lect. Notes in Math., 1530 (1992), 30–55. [45] A. Visintin, Models of Phase Transitions, Birkhauser, Boston, 1996. [46] F. Yi, A note on the classical solution of a two dimensional superconductor free boundary problem, European J. Appl. Math., 7 (1996), 113–118. [47] F. Yi and J. Liu, Vanishing specific heat for the classical solutions of a multidimensional Stefan problem with kinetic condition, Quart. Appl. Math., 57 (1999), 661–672.

Variational and Dynamic Problems for the Ginzburg-Landau Functional Halil Mete Soner Department of Mathematics Ko¸c University RumeliFener Yolu Sariyer 80910 Istanbul, Turkey [email protected]

Table of Contents 1. Introduction 2. Energy Lower Bounds a) Degree and Jacobian b) Covering argument c) Main lower bound 3. Jacobian and the GL Energy a) Jacobian estimate b) Compactness in two dimensions 4. Gamma Limit a) Functions of BnV b) Gamma limit of I
5. Compactness in Higher Dimensions 6. Dynamic Problems: Evolution of Vortex Filaments a) Energy identities b) Mean curvature flow and the distance function c) Convergence 7. References

1 Introduction In these notes, I survey several results on the Ginzburg-Landau functional
 1 E
(u) dx, E
(u) := |∇u|2 + [1 − |u|2 ]2 , Iˆ
(u) := 2 4 U where U is an open, bounded subset of IRn with smooth boundary, and  > 0 is a small parameter.


Partially supported by the National Science Foundation grant DMS-98-17525 and a NATO grant CRG 971561

L. Ambrosio et al.: LNM 1812, P. Colli and J.F. Rodrigues (Eds.), pp. 177–233, 2003. c Springer-Verlag Berlin Heidelberg 2003


178

Halil Mete Soner

This functional is a simpler version of the Ginzburg-Landau functional for superconductivity. The model for superconductivity has two fields; the complex valued order parameter u and the vector valued magnetic potential A. The functional Iˆ
is obtained by setting A to zero and by appropriate scaling. In superconductivity, the length of the order parameter |u| is proportional to the density of superconducting electrons. Hence the zeroes of u, called vortices, are the places where superconductivity is lost. The parameter  corresponds to the inverse of the Ginzburg-Landau parameter κ and a number of type II superconducting materials have large κ justifying the asymptotic regime considered in the notes. The lecture notes of Rubinstein [23] provides a good introduction. The starting point of these notes is the seminal work of Bethuel, Brezis and H´elein [6] which gives a detailed asymptotic description of the minimizers u
of Iˆ
as  tends to zero when U ⊂ IR2 . To briefly describe this result, let u
be the minimizer of Iˆ
among all functions u satisfying a given boundary data u = g. Since as  tends to zero the second term in E
forces the solution to have length one, it is natural to assume that g takes values in the unit circle S 1 . In the complex notation, g admits the representation g = eiφ for some possibly multi-valued function φ. Then, the boundary condition is u(x) = g(x) = eiφ(x) ,

x ∈ ∂U.

(1)

The degree of g around the origin (or the winding number) is an important parameter. Using the local representation g = eiφ and the fact that U ⊂ IR2 , the degree can be computed by the following formula
1 deg(g; ∂U ) = ∇φ · t dH1 (x), (2) 2π ∂U  where t is the unit tangent and ∂U · · · dH1 is the line integral around ∂U . If the boundary data g has zero degree, then φ is single valued. It is then relatively straightforward to show that u
converges strongly to the smooth function u = eiϕ where ϕ is unique harmonic function in U satisfying the boundary condition ϕ = φ. Hence, the interesting the case is d := deg(g; ∂U ) = 0. Then, by topological considerations, any function satisfying the boundary and in particular the minimizer u
must have d zeroes. This fact makes the problem interesting as each zero carries an energy of at least π ln(1/). To see this consider the problem with U = BR := {|x| < R}, and g(x) = eiN θ where θ(x) is the angle between x and the x-axis and N is an integer. Consider a test function v
(x) = f (|x|/) eiN θ , with some smooth, positive function f satisfying f (0) = 0, f (r) = 1 for all r ≥ 1. By calculus,

Variational and Dynamic Problems for the Ginzburg-Landau Functional

179

Iˆ
(v
) = N 2 π ln(1/) + O(1). The N 2 term indicates that it is better to have N distinct zeroes of degree one, instead of less zeroes with higher degree. Hence the minimizer u
is expected to have d distinct zeroes a
:= (a
1 , . . . , a
d ), again called vortices, each having degree one. The minimum energy Iˆ
(u
) behaves like dπ ln(1/) and the asymptotic behavior of u
is determined by the location of the zeroes a
. Bethuel, Brezis and H´elein, obtained the location of the zeroes by calculating the next term in the minimum energy, which they call the renormalized energy. The renormalized energy W (a
) is a function of the location of the zeroes, and it has a representation in terms of the solution of the Laplace equation with point sources at a
, or equivalently the canoniacl hramonic as defined in [6]. Then, the minimum energy has the form Iˆ
(u
) = d π ln(1/) + W (a
) + o(1).

(3)

In view of this expansion of the minimum energy, it is clear that a
converges to a minimum of the renormalized energy W . Once the location of the zeroes is determined it is possible to calculate the limit of u
. We refer to [6] for more information. The chief difference between the problem considered here and the model for superconductivity is the boundary condition. In superconductivity, Neumann condition is given and the vortices are formed by an exogenous forcing term which is the applied magnetic field. While in the above problem vortices created by the Dirichlet data. For this reason, local results which do not refer to a particular boundary condition are more useful for our understanding of the model for superconductivity (we refer to the recent paper of Sandier and Serfaty [27], and the references therein for infromation on mathematical results on the model for superconductivity.) The Gamma limit is such a result. For the scalar valued functions, the Gamma limit of the Ginzburg-Landau functional, with a different rescaling, is proved by Modica and Mortola [21, 21], and by Modica [20]. A brief definition of the Gamma limit is given in §4. In view of the above calculations, we consider the Gamma limit of the rescaled Ginzburg-Landau functional
1 Iˆ
(u)
= E
(u) dx. I (u) := ln(1/) ln(1/) U In §4 below we will show that the Gamma limit of I
is equal to  |Ju|(U ), if u ∈ B2V (U ; S 1 ), I(u) := +∞, otherwise, where Ju is the Jacobian of u and weakly it is given by (see §2.1) Ju :=

1 ∇ × j(u), 2

j(u) := u × Du = det(u; Du),

180

Halil Mete Soner

S 1 is the unit circle, and B2V is the set of all functions whose weak Jacobian is a Radon measure. The weak definition of the Jacobian in higher dimensions is discussed in §5, and BnV with a general n is defined properly in §4.1. This class of functions and its properties are studied in the two papers of the author with Jerrard [14, 15]. The set BnV , called functions of bounded n variations is related to the classical BV space and to the Cartesian currents of Giaquinta, Modica and Soucek [10, 11]. It is shown in [14] and also in §4.1 below that for u ∈ B2V (U ; S 1 ), the Jacobian has a special structure

Ju = π ki δai , i

for some points {ai } ⊂ U and integers ki . Here δai is the Dirac measure located at ai . This is interpreted as encoding the location and the topological singularities (or zeroes) of u. Moreover, for u ∈ B2V (U ; S 1 ),

|Ju|(U ) = |ki |. i

Hence, the Gamma limit I(u) counts the zeroes of u with multiplicity. This Gamma limit is proved in several steps. The first step is a “local” energy lower bound of the form
E
(u) dx ≥ π ln(1/) deg(u; ∂U ) − C, U

for some constant C. There are problems with this estimate as it is stated. The difficulty comes from the possible zeroes of u near or on ∂U . We will prove two such results, Theorem 2.1, and Theorem 2.2. They are local in nature, especially the second one. The proof technique is an elegant covering argument of Jerrard [12]. To explain this method clearly, we first prove it under slightly restrictive set of assumptions first to prove Theorem 2.1. We then modify this technique to obtain the sharper lower bound Theorem 2.2. A corollary of this lower bound is a compactness result for the Jacobian. This estimate bounds the Jacobian by the Ginzburg-Landau energy, and yields a compactness result for the Jacobian. Indeed for any sequence u
satisfying sup I
(u
) < ∞,


the Jacobians Ju
are compact in dual (C 0,α )∗ norm for every α > 0. Hence, on a subsequence Ju
converge to a distribution J not in a norm slightly weaker than the weak∗ topology of Radon measures. Although this convergence is not in the space of measures, we will show that the resulting distribution J is indeed a measure of a special form. For B2V (U ; IR2 ) with U ⊂ IRm with m = 2, this is proved in §3.2, and for m ≥ 3 it is stated in §5.

Variational and Dynamic Problems for the Ginzburg-Landau Functional

181

Then the Gamma limit is proved in §4 as a result of the lower bound and the compactness of the Jacobian. The compactness of the Jacobian is also a useful tool in the analysis of the dynamic problems. To motivate the study of the evolution problems in this context, let us consider an experiment in superconductivity. In this experiment the vortices are formed by an external magnetic field. Then the magnetic field is turned off and the material turns back to superconducting state. To understand these transition from the vortex state to the superconducting state, both the parabolic u [1 − |u|2 ], 2

ut − ∆u =

t > 0, x ∈ IRn ,

(4)

and the Schr¨ odinger i ut − ∆u =

u [1 − |u|2 ], 2

t > 0, x ∈ IRn ,

(5)

equations are proposed. The question then is to obtain evolution of the vortices that are forced into the system via the initial data. Mathematically this is achieved by studying the small  asymptotics of the above equations for an initial data u
0 which contains several vortices with degree ±1. Asymptotic expansion techniques are used by Neu, Rubinstein and E to derive these equations; see the lecture notes of Rubinstein [23]. Since (4) is the gradient flow for Iˆ
, in view of the expansion (3), as  tends to zero, the vortices should satisfy the gradient flow for the potential W . Indeed, let a(t) = (a1 (t), . . . , aN (t)) be the limit of vortices a
(t ln(1/)). Then, d a(t) = −∇W (a(t)), dt

(6)

in the case of (4). Note that we need to speed up the dynamics by a factor of ln(1/). For the Schr¨ odinger equation, in the original time scale, we get the Hamiltonian dynamics. These results are rigorously proved in several papers. For the parabolic flow, in [18] Lin proved that the speed of vortices is 1/ ln(1/) when the vortices all have same sign. The mixed vortex case, which is the relevant one in the experiment outline above, was proved in [16] by Jerrard and the author. First rigorous derivation of the vortex equation is also given in [16]. For U = IR2 , an explicit form of (6) is avaliable N

(ak (t) − aj (t)) d ak (t) = 2 dj dk , dt |ak (t) − aj (t)|2 j=1

where dk is the degree of u
around ak for small , and by hypothesis dk is equal to ±1. Note that the solutions of the above equation behaves like charged particles with a logarithmic potential; vortices with same degree expel each other while ones with opposite degree attract with a force proportional to the inverse of the distance between the vortices.

182

Halil Mete Soner

In IRn , the set of zeroes of u
is a codimension two set, as  tends to zero we obtain geometric equations for these sets; called vortex lines in IR3 . As expected from results on scalar version of (4), the limiting vortex line moves by mean curvature flow. We refer to [23, 24] and the references therein for the formal derivation of these equations. First rigorous results for the vector Ginzburg-Landau equation are [17] and [4]. In §6 we prove the convergence when there exists a smooth solution {Γt }t∈[0,T ] of the codimension two mean curvature flow. The main idea set forward in [17] is to consider the limiting measures µt := weak∗ limit of µ
t , where µ
t (V ) :=

1 | ln 1/|


E
(u
(t, x)) dx. V

In Theorem 6.1, under appropriate assumptions on the initial data, we will show that µt ≥ π Hn−2 Γt , support µt = Γt , where Hn−2 Γt is the Hausdorff measure restricted to Γt , i.e., it is the surface area measure on the surface Γt . Moreover, the limit Jt of the Jacobians Ju
(t, ·) satisfies |Jt | = π H1 Γt . (7) To prove the inclusion support (µt ) ⊂ Γt , we use the energy identities and a Pohazaev type inequality as in [17]. The idea is to estimate the time derivative of
α(t) := η(t, x) µt (dx), IRn

when the test function η is the square distance function of Γt . Since {Γt }t∈[0,T ] solves the mean curvature flow, the square distance function η satisfies ∇ηt = ∇∆η,

on Γt .

We use this and the other properties of the square distance function to prove that α(t) = 0 for t ∈ [0, T ]. This proves the inclusion support (µt ) ⊂ Γt . The opposite inclusion is proved by studying the Jacobian. In view of the energy estimate

T




2 E (u (T, x)) dx + |ut (t, x)| dx dt = E
(u
(0, x)) dx, IRn

0

IRn

IRn

our compactness result implies that Jt
:= Ju
(u
(t, ·)) is compact. Let Jt be a limit of Jt
. Then, by the previous inclusion we know that the support of Jt is in Γt . Moreover, by the weak formulation of the Jacobian (see §2.1), the Jacobian is divergence free. Since Γt is smooth manifold with no boundary,

Variational and Dynamic Problems for the Ginzburg-Landau Functional

183

this implies that the density of the Jacobian on Γt is constant. We then show that this constant is equal to π for all t ∈ [0, T ], proving (7). Also, in view of the compactness result, Theorem 5.2, the energy measure dominates the Jacobian measure. Hence, the support of µt is equal to Γt . Acknowledgments. I would like to thank the organizers, Professors Colli and Rodrigues, of the CIM/CIME Euro-Summer School for giving me the oppurtunity to put together several results obtained in different papers into these lecture notes, and also for a very productive Summer School.

2 Energy Lower Bounds In this section we prove an energy lower bound in terms of the topological degree. This bound is local in nature and is a key step in the proof of the Gamma limit as well as the dynamical properties of the vortices. Local energy lower bounds were proved by covering arguments by Jerrard [12] and Sandier [25]. Here, we follow the technique developed by Jerrard to prove these estimates. In the next subsection, we give a brief and a formal discussion of this technique. Then we will give the precise statement and the proof of the lower bound. Let U is a bounded open subset of IR2 , and u ∈ H 1 (U ; IR2 ) is a function that we have assumed (without loss of generality) to be smooth. The goal is to find an energy estimate of the form
E
(u) dx ≥ π ln(1/) deg(u; ∂U ) − C, U

for some constant C, independent of  and u. We want this to hold for all u ∈ H 1 (U ; IR2 ) and  ∈ (0, 1]. However, there are problems with this estimate as it is stated. The function u may have a zero on the boundary of U . Then, the degree of u around ∂U is not even be defined. Also, when u has a zero very close to the boundary, most of the energy could be outside the domain U . These possibilities indicate that we have make either an assumption about the boundary behavior of u, or to modify the statement of the lower bound. The latter is better suited for the later use of these estimate as it makes the lower bound a “local” one. However the statement of this local lower bound is rather technical. To explain the main idea we first outline the proof under assumptions on the boundary behavior. Remarkably, the proof technique of this “easier” lower bound carried over the more technical one with very little change. We start with a brief discussion of the degree. 2.1 Degree and Jacobian In our arguments, we will use the degree and the Jacobian repeatedly. For that reason we recall their definition.

184

Halil Mete Soner

Let V ⊂ U ⊂ IR2 and |u| = 0 on ∂V . Then, u admits a local representation u = |u|eiφ on ∂V , and the degree of u around ∂V is given by
1 deg(u; ∂V ) = ∇φ · t dH1 (x), 2π ∂V  where as in the Introduction, t is the unit tangent and ∂V · · · dH1 is the line integral around ∂V . For future reference, we note that deg(u; ∂V ) = deg(u/|u|; ∂V ). The Jacobian Ju of u satisfies Ju =

1 ∇ × j(u), 2

where j(u) = u × Du = det(u; Du). Hence by the Stokes’ theorem

Ju dx = V

j(u) · t dH1 .

(8)

∂V

Generalizations to the case when U ⊂ IRn will be discussed later. For u = |u|eiϕ , we directly calculate that j(u) = |u|2 ∇ϕ. Hence, ∇ϕ = j(v) = j(u)/|u|2 ,

v = u/|u|,

and by the Stokes’ theorem

j(u) · t 1 1 1 dH = j(v) · t dH1 , deg(u; ∂V ) = 2π ∂V |u|2 2π ∂V

(9)

for all u which do not vanish on ∂V . 2.2 Covering argument In this subsection we outline a covering technique developed by Jerrard to prove energy lower bounds [12]. Similar techniques were also used by Sandier [25]. To simplify the presentation, we assume that |u(x)| >

1 , whenever x ∈ Ur0 , 2

Ur0 := { x ∈ U | dist(x, ∂U ) ≥ r0 }, (10)

for some constant r0 > 0. We also assume that deg(u; ∂U ) = 0.

Variational and Dynamic Problems for the Ginzburg-Landau Functional

185

Theorem 2.1 (Jerrard [12]). There exists a constant C, such that for all  ∈ (0, 1], and for all u ∈ H 1 satisfying above assumptions,
E
(u) dx ≥ π ln(r0 /) − C. U

A more general result which do not assume (10) will be proved later in this section. We introduce some notation and definitions, taken from [12]. We let S denote the set on which |u| is small, that is, {x ∈ U : |u(x)| ≤ 1/2}.

(11)

We define the essential part SE of S to be SE := ∪{components Si of S : deg(u; ∂Si ) = 0}.

(12)

For any subset V ⊂ U such that ∂V ∩ SE = ∅, we define the generalized degree

dg(u; ∂V ) := {deg(u; ∂Si ) | components Si of SE such that Si ⊂ V } . (13) Notice that if u is nonzero on the boundary of V , the generalized degree agrees with the degree of u around ∂V . Hence the generalization of the degree is only relevant when u has zeroes on ∂V . But in this case, we could remove these zeroes by slightly modifying u and with small change in the energy. Hence, in view of the Ginzburg-Landau energy these zeroes are removable and this justifies the definition of SE and the generalized degree. In view of (10), Ur0 ∩ SE = ∅, and by the definition of the generalized degree and the assumption that the degree of u around ∂U is nonzero, dg(u; ∂Ur ) = deg(u; ∂Ur ) = deg(u; ∂U ) = 0,

∀ r ∈ (0, r0 ].

Our strategy for proving Theorem 2.1 will be to find a collection of balls with a good lower bound for the Ginzburg-Landau energy on each ball. We then show that the sum of the radii of the balls is bounded below by r0 /2, hence obtaining a lower bound for the total Ginzburg-Landau energy in terms of this quantity. This will be done in several steps.

186

Halil Mete Soner

1. First cover of SE . We find the collection of balls by starting from an initial collection of
, then letting these balls grow by expanding them small balls that cover SE and combining them. The first step is thus to establish the existence of the initial collection of small balls. This is the content of Proposition 2.1. There is a collection of closed, pairwise disjoint balls {Bi∗ }ki=1 with radii ri∗ such that SE ⊂ ∪ki=1 Bi∗ ,

(14)

ri∗ ≥ 

(15)

∀i,


E
(u)dx ≥

Bi∗ ∩U

c0 ∗ r .  i

(16)

Proof. This is proved in [12]. Let {Si }{i=1,...,N } be the disjoint components of SE . Choose xi ∈ Si for each i, and set ρi := inf{r > 0 : ∂Br (xi ) ∩ Si = ∅},

ri := max{ ; ρi }.

Set Bi := Bri (xi ) so that in view of the definition of ri Si ⊂ Bi ∩ U. Note that |u| = 1/2 on ∂Si , and E
(u) ≥

1 |Du|2 ≥ |Ju|. 2

By (8) and (9),



   E
(u) dx ≥  Ju dx Bi ∩U 
Si    1  ≥ j(u) · tdH  ∂Si

1 ≥ |deg(u; ∂Si )| 2π 1 . ≥ 2π Moreover, ∂Br (xi ) ∩ Si = ∅,

∀ r ∈ (0, ρi ].

This means that for every r ∈ (0, ρi ], there is a point x∗ ∈ ∂Br (xi ) such that |u(x∗ )| < 1/2. Due to the potential term (1 − |u|2 )2 /42 term in the energy E
, near x∗ , is large and in Lemma 2.3 below we will show that

Variational and Dynamic Problems for the Ginzburg-Landau Functional


E
(u) dH1 ≥ C ∂Br (xi )∩U

1 , 

187

∀ r ∈ [, ρi ],

for some constant C. Assume that ri ≥ , and integrate this estimate over [, ri ]. The result is
(ri − )+ . E
(u) dx ≥ C  Bi ∩U Combining the two estimates,
C ri (ri − )+ ; 1}≥ . E
(u) dx ≥ C max{  2  Bi ∩U The balls constructed above may not be disjoint. If two or more of these balls intersect, they can be combined into larger balls, relabeling as necessary. One can use the Besicovitch Covering Theorem to control the overlap and show that the larger balls still satisfy (16). The details of this argument appear in [12]. 2 2. Annulus estimate. In the previous step, we did not attempt to make the covering as large as possible. In particular, they could be off the size  and when we add them we will not get the desired energy estimate. In this step, we obtain an estimate which will be used when we extend the balls in our covering. Suppose that x∗ ∈ U , and  ≤ r0 < r1 satisfy [Br1 (x∗ ) \ Br0 (x∗ )] ∩ SE = ∅, and

dg(u; ∂Br1 (x∗ )) = 0.

Then, for all r ∈ [r0 , r1 ], dg(u; ∂Br (x∗ )) = dg(u; ∂Br1 (x∗ )) = 0. Set






λ (r) = min m∈[0,1]

 m2 π (1 − m)2 + , r c0 




r

λ
(s) ∧




Λ (r) := 0

c1 ds, 

(17)

for certain constants c0 , c1 whose choice is discussed below. Lemma 2.1. There are constants c0 and c1 so that
E
(u)dx ≥ [Λ
(r1 ) − Λ
(r0 )]. Br1 (x0 )\Br0 (x0 )

(18)

188

Halil Mete Soner

Proof. This is Proposition 3.2, [12]. The key estimate is
E
(u) dH1 ≥ λ
(r), ∀r ∈ [r0 , r1 ].

(19)

∂Br (x∗ )∩U

We then obtain (18) by integrating (19) over r ∈ [r0 , r1 ]. Set m := min∗ { |u(x)| }. ∂Br (x )

Then, as in Lemma 2.3 below, we can prove that
(1 − m)2 . E
(u) dH1 ≥ c0  ∂Br (x∗ )∩U For u = |u|ei ϕ, E
(u) =

1 1 2 |u|2 |∇ϕ|2 + |∇|u|| . 2 2

Since |j(u)| = |u|2 |∇ϕ|, E
(u) =

1 |j(u)|2 1 1 m2 |j(u/|u|)|2 . + |∇|u||2 ≥ |u|2 |j(u/|u|)|2 ≥ 2 2 |u| 2 2 2

Suppose that m > 0. Then, for r ∈ [r0 , r1 ], deg(u; ∂[Br∗ ∩ U ]) = dg(u; ∂[Br∗ ∩ U ]) = 0, and

m2 E
(u) dH1 ≥ |j(u/|u|)|2 dH1 2 ∂[Br (x∗ )∩U ] ∂[Br (x∗ )∩U ] 
2  m2   ≥ j(u/|u|) dH1    4πr  ∂[Br (x∗ )∩U ] π m2 2 |deg(u; ∂[Br (x∗ ) ∩ U ]| r π m2 . ≥ r =

2

Combining the two preceeding estimates we obtain (19). 3. Properties of λ
.

The following elementary estimates are proved in Propositions 3.1 and 3.2 in [12] Λ
(r1 + r2 ) ≤ Λ
(r1 ) + Λ
(r2 ) , s →

1
Λ (s) s

is non increasing ,

1
c1 Λ (s) ≤ s 

(20) ∀s

(21)

Variational and Dynamic Problems for the Ginzburg-Landau Functional

189

and Λ
(r) ≥ π ln(r/) − c2 for some constant c2 . Also, clearly, λ
(r) ≤ π/r, and therefore, by redefining c2 if necessary, |Λ
(r) − π ln(r/) | ≤ c2

∀r≥.

(22)

4. Amalgamation. Our next result is Lemma 3.1 in [12]. It is used below when we allow the small balls to grow and merge, to form large balls. For the sake of completeness, here we state it and give its short proof. Lemma 2.2. Given any finite collection of closed balls in IRk , say {Ci }N i=1 , ˜ of pairwise disjoint balls such that we can find a collection {C˜i }N i=1 2N i=1

" ˜i Cj ⊂C

˜ ≤ N, N

Ci ⊂

2N˜ i=1

C˜i ,

diamCj = diam C˜i ,

with strict inequality unless {Ci }N i=1

is pairwise disjoint.

Proof. Replace pairs of intersecting balls Ci , Cj by larger single balls C˜ such that Ci ∪ Cj ⊂ C˜ and diam C˜ = diam Ci + diam Cj , continuing until a pairwise disjoint collection is reached. This collection has the stated properties. 2 5. Cover of SE . We next show that, starting from the initial collection of balls, we can let them grow in such a way that each ball continues to satisfy a good lower bound. As above let {Bi∗ } denote the balls found in Proposition 2.1, with radii ∗ ri and generalized degree d∗i := dg(u; ∂[Bi∗ ∩ U ]). Define σ ∗ := min{ ri∗ | d∗i = 0 } . The idea is to extend the balls with the smallest radius with nonzero degree until they hit each other or the boundary of U . When they hit each other we use the amalgamation lemma and continue the process until one of the balls with nonzero degree hits the boundary of U . However, here we follow the presentation of Sandier and Serfaty [27]. Although this is slightly more technical than the outline procedure, it extends very easily to more general situations.

190

Halil Mete Soner

Proposition 2.2. For every σ ≥ σ ∗ , there exists a collection of disjoint, k(σ) closed balls B(σ) = {Bkσ }k=1 satisfying rkσ ≥ , SE ⊂ ∪k Bkσ ,
U ∩Bkσ

rkσ ≥ σ

E
(u) dx ≥

(23)

rkσ
Λ (σ) , σ

whenever Bkσ ∩ ∂U = ∅ , and dσk = 0,

(24) (25)

where rkσ is the radius and dσk = dg(u; ∂[Bkσ ∩ U ]) is the generalized degree. Proof. Let C be the set of all σ ≥ σ ∗ for which such a collection exists. 1. We first claim that σ ∗ ∈ C. Indeed {Bk∗ } be the collection of balls constructed in Proposition 2.1. Set B(σ ∗ ) := {Bk∗ }. The definition 17 of Λ
easily implies that Λ
(σ)/σ ≤ c1 / forall σ, so Proposition 2.1 implies that this collection satisfies (23) and (24). Also, (25) is satisfied due to the definition of σ ∗ . 2. In step we will show that C is closed. Let {σ n }n be a sequence in C and suppose that σ n converges to σ0 as n tends to infinity. Since the balls are disjoint, and their radii are at least , the total number of balls k(σn ) is uniformly bounded in n. Therefore by passing to a subsequence we may assume that k(σn ) is equal to a constant k0 independent of n. By passing to a further subsequence, we may assume that the radii rkσn and the centers aσk n converge to rk0 and a0k , respectively, for each k ≤ k0 . Let Bk,0 be the closed ball centered at a0k with radius rk0 . It is clear that this collection of balls satisfies (23), (24), and (25). If the balls are disjoint , we 0 set B(σ0 ) := {Bk,0 }kk=1 . If they are not disjoint, we apply the amalgamation process outlined in Lemma 2.2. Let {Bjσ0 } be the resulting balls and rjσ0 be their radius. Then, by Lemma 2.2,

rjσ0 = rk,0 . (26) σ

Bk,0 ⊂Bj 0

Since {Bk,0 } satisfies (24), this implies that



E
(u) dx ≥ σ

U ∩Bj 0

σ Bk,0 ⊂Bj 0

≥



σ

Bk,0 ⊂Bj 0

=

E
(u) dx

U ∩Bk,0

rk,0
Λ (σ0 ) σ0

rjσ0
Λ (σ0 ) . σ0

Hence, B(σ0 ) := {Bjσ0 } satisfies (24). Moreover,

Variational and Dynamic Problems for the Ginzburg-Landau Functional

     

 σ    d 0  =  d k,0  ≤ j    Bk,0 ⊂Bjσ0



191

|dk,0 |.

σ Bk,0 ⊂Bj 0

Hence, if Bjσ0 ∩ ∂U = ∅ and dσj 0 = 0, then Bk,0 ∩ ∂U = ∅ for all Bk,0 ⊂ Bjσ0 and at least one dk∗ ,0 = 0. Since Bk∗ ,0 satisfies (25), rjσ0 ≥ rk∗ ,0 ≥ σ0 . This implies that the balls in the collection B(σ0 ) satisfy (25). 3. Suppose that σ1 ∈ C. We will show that there is δ > 0 such that [σ1 , σ1 + δ] ⊂ C. Indeed, let K1 := { k | Bkσ1 ∩ ∂U = ∅ and dσk 1 = 0 }, and set s1 := min { rkσ1 }. k∈K1

By (25), σ1 ≤ s1 . If this inequality is strict, we set B(σ) = B(σ1 ) for all σ ∈ [σ1 , s1 ]. It is clear that this collection of balls satisfies (23), and (25). Also (24) follows from (21). So let us assume that s1 = σ1 , and let K2 := { k ∈ K1 | s1 = rkσ1 }. For σ ≥ σ1 , set rkσ

:=

 σ1  rk ,

if k ∈ K2 ,



if k ∈ K2 .

σ σ1

rkσ1 ,

Let Bkσ be the closed ball with radius rkσ with the same center as Bkσ1 and let B(σ) be the collection of these balls. Since {Bkσ1 }k are disjoint closed sets, there is δ1 > 0 such that for all σ ∈ [σ1 , σ1 + δ1 ] Bkσ ’s are disjoint and K1 (σ) := { k | Bkσ ∩ ∂U = ∅ and dσk = 0 } = K1 . Then, for k ∈ K2 ,

and for k ∈ K2 ,

r σ1 rkσ = k = 1, σ σ1 rkσ σ1 rkσ1 = . σ σ σ1

Since for k ∈ K2 , rkσ1 > σ1 , there is 0 < δ ≤ δ1 such that (25) is satisfied by the collection B(σ). Since rkσ ≥ rkσ1 , (23) is also satisfied. To verify (24), we observe that for k ∈ K2 , Bkσ = Bkσ1 and (24) is satisfied in light of (21). If, however, k ∈ K2 , then dσk = dσk 1 ,

rkσ = σ ,

(27)

192

Halil Mete Soner

and [Bkσ \ Bkσ1 ] ∩ S E = ∅ . Then by (18),



E
(u) dx =

Bkσ


σ

Bk 1 rkσ1

E
(u) dx +

σ

Bkσ \Bk 1

E
(u) dx

Λ
(σ1 ) + [Λ
(rkσ ) − Λ
(rkσ1 )] σ1 = Λ
(rkσ1 ) + [Λ
(rkσ ) − Λ
(rkσ1 )] = Λ
(rkσ ) rσ = k Λ
(σ) . σ ≥

Here we repeatedly used the identities (27) and the fact that Bkσ1 satisfies (24). Hence B(σ) also satisfies (24) for all σ ∈ [σ1 , σ1 + δ]. 4. We have shown that C is closed set including σ ∗ and for every σ ∈ C, there exists δ > 0 such that [σ, σ + δ] ⊂ C. Hence, C = [σ ∗ , ∞). 2 6. Proof of Theorem 2.1 Let σ ∗ be as in the previous Lemma and let r0 be as in (10). 1. First suppose that r0 < 2σ ∗ . The opposite inequality will be treated later in the proof. Consider the balls {Bk∗ } constructed in Proposition 2.1. Set K ∗ := { k | and d∗k = 0 }. Recall that d∗k = dg(u; ∂[Bk∗ ∩ U ]). Since

0 = deg(u; ∂U ) = dg(u; Bk∗ ), k

K ∗ is nonempty. Then, by (16) and the definition of σ ∗ ,





c1
rk∗ E (u) dx ≥ E
(u) dx ≥ ∗  U U ∩B k k k r0 r0 c1 σ ∗

|K ∗ | ≥ c1 ≥ 1 ≥ c1  2 2 k∈K ∗  
r0 . ≥Λ 2 In view of (22), this gives the desired lower bound. 2. We now assume that r0 ≥ 2σ ∗ . Set σ ¯ := r0 /2 and consider the collection of balls B(¯ σ ) provided by Proposition 2.2. Let

Variational and Dynamic Problems for the Ginzburg-Landau Functional

193

¯ := { k | dσk¯ = 0 }. K ¯ is nonempty. We claim that Since deg(u; ∂U ) = 0, K ¯= rkσ¯ ≥ σ

r0 , 2

¯ ∀ k ∈ K.

¯ then this follows from (25). Suppose Indeed if Bkσ¯ ∩ ∂U = ∅ for some k ∈ K, σ ¯ ¯ that Bk ∩ ∂U = ∅ for some k ∈ K. Since dσk¯ = 0, Bkσ¯ contains a zero of u, and by (10), Bkσ¯ ∩ [U \ Ur0 ] = ∅. Since, by assumption, Bkσ¯ ∩ ∂U = ∅, this implies that the diameter of Bkσ¯ is at least r0 , proving the claim. 3. By the previous step

r0 rkσ¯ ≥ σ ¯= . 2 ¯ k∈K

Hence by (24),
E
(u) dx ≥ U


¯ k∈K

≥



¯ U ∩Bkσ

rkσ¯

¯ k∈K


E
(u) dx

Λ
(¯ σ) σ ¯

σ) . ≥ Λ (¯ 2 2.3 Main lower bound In this section, we prove a generalization of the lower bound. This sharper lower bound is taken from [13] and its proof is very similar to that of Theorem 2.1. Here we repeat the arguments of the previous section with minor modifications for the sake of completeness. Let U be a bounded open subset of IR2 , and u ∈ H 1 (U ; IR2 ) be a function that we have assumed (without loss of generality) to be smooth. In addition, φ is a nonnegative Lipschitz test function that vanishes on ∂U . Throughout this section we will use the notation T = φ∞ = max φ(x). U

(28)

Given φ ∈ Cc0,1 (U ) we use the notation Reg(φ) for the set   t ∈ [0, T ] : ∂Ω(t) = φ−1 (t), ∂Ω(t) is rectifiable, H1 (∂Ω(t)) < ∞ . (29)

194

Halil Mete Soner

The coarea formula implies that Reg(φ) is a set of full measure. For every t ∈ Reg(φ), ∂Ω(t) is a union of finite Jordan curves Γi (t), i.e., ∂Ω(t) = ∪i Γi (t) ,

∀ t ∈ Reg(φ).

In particular, this holds for almost every t. For t ∈ Reg(φ) we define   Γ (t) = ∪ components Γi (t) of ∂Ω(t) | min |u(x)| > 1/2 .

(30)

We also define γ(t) = ∂Ω(t) \ Γ (t),  γ(t) = ∪ components Γi (t) of ∂Ω(t) |

(31)

x∈Γi (t)

 min |u(x)| ≤ 1/2 .

x∈Γi (t)

When we want to indicate explicitly the dependence of Γ (t) on φ and u, we will write Γφ,u (t). We will also use the notation t
:= ∇φ∞ .

(32)

For any positive integer d, let Dd
:= {t ∈ Reg(φ) : t ≥ t
, Γ (t) is nonempty, and |deg(u; Γ (t))| ≥ d} . (33) The main result of this section is the following theorem. We follow very closely arguments introduced in [12]. Theorem 2.2 (Jerrard & Soner [13]). If u : U → IR2 is a smooth function and φ is a nonnegative Lipschitz function such that φ = 0 on ∂U , then for any positive integer d,  
|Dd
| . E
(u) ≥ dΛ
2d∇φ∞ spt(φ) For any t2 > t1 the ratio (t2 −t1 )/∇φ∞ is a lower bound for the distance between ∂Ω(t2 ) and ∂Ω(t1 ). This explains the role of ∇φ∞ in the estimate. Similar results were proven in [12] under more or less the assumption that Dd
is an interval; and in [7] in the case d = 1. Related results have also appeared in Sandier [26]. Note that the case covered in the statement of the theorem, {x : φ(x) > 0} ⊂ U , can be reduced to the case {x : φ(x) > 0} = U , if we replace U ˜ := {x ∈ U : φ(x) > 0}. So we will henceforth assume for notational by U ¯. simplicity that this holds, so that spt(φ) = U For the proof of Theorem 2.2, we define

Variational and Dynamic Problems for the Ginzburg-Landau Functional
SE := ∪{components Si of SE : Si ⊂ Ω(t
)}.

195

(34)

If x ∈ Ω(t
) and y ∈ ∂U , then |x − y| ∇φ∞ ≥ |φ(x) − φ(y)| = |φ(x)| ≥ t
= ∇φ∞ . In particular, dist(x, ∂U ) ≥ 


for all x ∈ SE .

(35)

Note also that if V ⊂ Ω(t
) and ∂V ∩ SE = ∅, then


dg(u; ∂V ) := {deg(u; ∂Si ) | components Si of SE such that Si ⊂⊂ V } .
In other words, for such sets V we can ignore SE \ SE when computing dg(u; ∂V ). In the proof of Theorem 2.2 below we will always be concerned with subsets V ⊂ Ω(t
), so this will always be the case. Our strategy for proving Theorem 2.2 is very similar to the method we used to prove Theorem 2.1. We will first find a collection of balls such we have a good lower bound for the Ginzburg-Landau energy on each ball. We then show that the sum of the radii of the balls is bounded below by |Dd
|/(2∇φ∞ ), hence obtaining a lower bound for the total GinzburgLandau energy in terms of this quantity. We start with a technical step.

1. If γ(t) = ∅. In this step, we prove an estimate in the case when one of the level sets intersects with the zero set. In this case, |u| falls below 1/2 on γ(t) and we expect the Ginzburg-Landau energy to be large on γ(t). The following technical lemma proves this under the assumption that γ(t) is not too small. Lemma 2.3. Suppose that H1 (γ(t)) ≥  . Then


E
(u) dH1 ≥ ∂Ω(t)

1 . 25

(36)

Proof. This is very similar to Lemma 2.3 in [12]. Fix a connected component Γi (t) of γ(t) and set ρ := |u| and
1 γi := |∇ρ|2 dH1 . Γi (t) 2 By the definition (31) of γ(t) there is a point xmin ∈ Γi (t) such that ρ(xmin ) ≤ 1/2. Parametrize Γi (t) by arclength so that

196

Halil Mete Soner

Γi (t) = { x(s) | s ∈ [0, Gi ]} ,

Gi := H1 (Γi (t)) ,

with xmin = x(0) = x(Gi ). Then since |x(s)| ˙ = 1,
s ρ(x(s)) = ρ(x(0)) + ∇ρ(x(r)) · x(r) ˙ dr 0


s 1/2 1 1/2 2 |∇ρ(x(r))| dr ≤ +s 2 0 1 √ 3 ≤ + γi s ≤ , 2 4 provided that s ≤ σi := [Gi ∧ 1/(16γi )]. Then, for s ∈ [0, σi ], (1 − ρ2 (x(s)))2 /4 ≥ 1/25. Therefore,

1
1 E (u) dH ≥ γi + (1 − ρ2 )2 dH1 2 4 Γi (t) Γi (t) σi ≥ γi + . 25 2 By calculus, γi +

σi Gi ∧ (1/4γi ) 1 = γi + ≥ 2 2 25  25  25


Thus

1 E (u) dH ≥ 25 Γi (t)


1





Gi ∧5 

Gi ∧5 

 .

 .

Since H1 (γ(t)) =

{i|Γi (t) is a component of γ(t)}

H1 (Γi (t)) =



Gi ≥  ,

i

we can sum over components Γi (t) of γ(t) to conclude that
1 . E
(u) dH1 ≥ 25 ∂Ω(t) 2 2. First Cover. This is very similar to Step 1 of the previous subsection. We find the collection of balls by starting from an initial collection of
small balls that cover SE , then letting these balls grow by expanding them and combining them. The first step is thus to establish the existence of the initial collection of small balls. This is the content of

Variational and Dynamic Problems for the Ginzburg-Landau Functional

197

Proposition 2.3. There is a collection of closed, pairwise disjoint balls {Bi∗ }ki=1 with radii ri∗ such that
⊂ ∪ki=1 Bi∗ , SE

(37)

ri∗ ≥ 

(38)


Bi∗ ∩U

E
(u)dx ≥

∀i, c0 ∗ r ≥ Λ
(ri∗ ).  i

(39)

This is essentially proved in the previous subsection; Proposition 2.1 and in Proposition 3.3 in [12]. Proposition 2.3 differs from Proposition 2.1 in that in the latter, SE ap
pears in place of SE in the counterpart of (37).
3. Cover of SE .

This step is similar to Step 5 of the previous section. Starting from the initial collection of balls, we can let them grow in such a way that each ball continues to satisfy a good lower bound. As above let {Bi∗ } denote the balls found in Proposition 2.3, with radii ∗ ri and generalized degree d∗i := dg(u; ∂[Bi∗ ∩ U ]). Define σ ∗ := min{

ri∗ | d∗i = 0 } . |d∗i |

Proposition 2.4. For every σ ≥ σ ∗ , there exists a collection of disjoint, k(σ) closed balls B(σ) = {Bkσ }k=1 satisfying rkσ ≥ ,
⊂ ∪k Bkσ , SE


U ∩Bkσ

rkσ ≥ σ|dσk |

E
(u) dx ≥

rkσ
Λ (σ) , σ

whenever Bkσ ∩ ∂U = ∅ ,

(40) (41) (42)

where rkσ is the radius and dσk is the generalized degree. The proof of this proposition is very similar to that of Proposition 2.2. The only difference is we consider the ratio rkσ /|dσk | to decide which balls to expand. Proof. Let C be the set of all σ ≥ σ ∗ for which such a collection exists. 1. We first claim that σ ∗ ∈ C. Indeed {Bk∗ } be the collection of balls constructed in Proposition 2.3. Set B(σ ∗ ) := {Bk∗ }. The definition (17) of Λ
easily implies that Λ
(σ)/σ ≤ c1 / forall σ, so Proposition 2.3 implies

198

Halil Mete Soner

that this collection satisfies (40) and (41). Also, (42) is satisfied due to the definition of σ ∗ . 2. In step we will show that C is closed. Let {σ n }n be a sequence in C and suppose that σ n converges to σ0 as n tends to infinity. Since the balls are disjoint, and their radii are at least , the total number of balls k(σn ) is uniformly bounded in n. Therefore by passing to a subsequence we may assume that k(σn ) is equal to a constant k0 independent of n. By passing to a further subsequence, we may assume that the radii rkσn and the centers aσk n converge to rk0 and a0k , respectively, for each k ≤ k0 . Let Bk,0 be the closed ball centered at a0k with radius rk0 . It is clear that this collection of balls satisfies (40), (41), and (42). If the balls are disjoint , we 0 set B(σ0 ) := {Bk,0 }kk=1 . If they are not disjoint, we apply the amalgamation process outlined in Lemma 2.2. Let {Bjσ0 } be the resulting balls and rjσ0 be their radius. Then, by Lemma 2.2,



rjσ0 = rk,0 ≥ σ0 |dk,0 |. (43) Bk,0 ⊂Bj

Bk,0 ⊂Bj

Since {Bk,0 } satisfies (41), this implies that




E (u) dx ≥ U ∩Bj

Bk,0 ⊂Bj

≥ ≥



Bk,0 ⊂Bj rjσ0


σ0

E
(u) dx

U ∩Bk,0

rk,0
Λ (σ0 ) σ0

Λ (σ0 ) .

Hence, B(σ0 ) := {Bjσ0 } satisfies (41). Moreover,    

   σ  d 0  =  dk,0  ≤ j   Bk,0 ⊂Bj



|dk,0 |,

Bk,0 ⊂Bj

and this together with (43) implies that the balls in the collection B(σ0 ) satisfy (42). 3. Suppose that σ1 ∈ C. We will show that there is δ > 0 such that [σ1 , σ1 + δ] ⊂ C. Indeed, let K1 be the set of indices k such that Bkσ1 ∩ ∂U = ∅ and set rkσ1 . s1 := min σ k∈K1 |dk 1 | By (42), σ1 ≤ s1 . If this inequality is strict, we set B(σ) = B(σ1 ) for all σ ∈ [σ1 , s1 ]. It is clear that this collection of balls satisfies (40), and (42). Also (41) follows from (21). So let us assume that s1 = σ1 , and let K2 ⊂ K1 be the indices k which minimize the ratio rkσ1 /dσk 1 . For σ ≥ σ1 , set

Variational and Dynamic Problems for the Ginzburg-Landau Functional

rkσ :=

 σ1  rk ,

if k ∈ K2 ,



if k ∈ K2 .

σ σ1

rkσ1 ,

199

Let Bkσ be the closed ball with radius rkσ with the same center as Bkσ1 and let B(σ) be the collection of these balls. Since {Bkσ1 }k are disjoint closed sets, there is δ1 > 0 such that for all σ ∈ [σ1 , σ1 + δ1 ] Bkσ ’s are disjoint and K1 (σ) := { k | Bkσ ∩ ∂U = ∅ } = K1 . Then, for k ∈ K2 ,

r σ1 rkσ = k = |dσk 1 | = |dσk | , σ σ1

and for k ∈ K2 ,

rkσ σ1 rkσ1 = . σ σ σ1

Since for k ∈ K2 , rkσ1 /σ1 > |dσk 1 |, there is 0 < δ ≤ δ1 such that (42) is satisfied by the collection B(σ). Since rkσ ≥ rkσ1 , (40) is also satisfied. To verify (41), we observe that for k ∈ K2 , Bkσ = Bkσ1 and (41) is satisfied in light of (21). If, however, k ∈ K2 , then rkσ = σ|dσk | ,

dσk = dσk 1 ,

(44)

and [Bkσ \ Bkσ1 ] ∩ S E = ∅ . Then by Lemma 2.1

E
(u) dx = Bkσ


σ

Bk 1 rkσ1

E
(u) dx +

σ

Bkσ \Bk 1

E
(u) dx

  σ1  rk rkσ
≥ −Λ Λ (σ1 ) + Λ σ1 |dσk | |dσk 1 |  σ1   σ1    σ  rk rk rk σ1 σ1


+ |dk | Λ −Λ = |dk | Λ |dσk 1 | |dσk | |dσk 1 |  σ  rk σ1
= |dk | Λ |dσk | rσ = k Λ
(σ) . σ 






|dσk 1 |




Here we repeatedly used the identities (44) and the fact that Bkσ1 satisfies (41). Hence B(σ) also satisfies (41) for all σ ∈ [σ1 , σ1 + δ]. 4. We have shown that C is closed set including σ ∗ and for every σ ∈ C, 2 there exists δ > 0 such that [σ, σ + δ] ⊂ C. Hence, C = [σ ∗ , ∞). We are now ready for the

200

Halil Mete Soner

Proof of Theorem 2.2. Set R := |Dd
|/(2∇φ∞ ) and σ ¯ := R/d. Let σ ∗
be as in the previous Lemma. We suppose that Dd is nonempty as there is nothing to prove otherwise . 1. First suppose that σ ¯ < σ ∗ . The opposite inequality will be treated later in the proof. Consider the balls {Bk∗ } constructed in Proposition 2.3. By (39) and the definition of σ ∗ ,




E (u) dx ≥ E
(u) dx U

U ∩Bk∗

k

≥

c1 R ∗ c1 σ ∗ ∗ rk∗ ≥ |dk | ≥ c1 |dk | .   d k

k

k

Let t0 ∈ Dd
. Then the definition (33) of Dd
implies that d ≤ |deg(u; Γ (t0 ))| and by definition, (30), |u| > 1/2 on Γ (t0 ). Hence d ≤ |dg(u; Γ (t0 ))|. Moreover, by (13) and (37),



d ≤ |dg(u; Γ (t0 ))| ≤ |d∗k | ≤ |d∗k | . {k : Bk∗ ∩Ω(t0 ) =∅}

k

Hence, by (21),


R d ≥ dΛ
E (u) dx ≥ c1 d


U



R d

 ,

which is what we needed to prove. σ) 2. We now assume that σ ¯ ≥ σ ∗ . Consider the collection of balls B(¯ provided by Proposition 2.4. Assume towards a contradiction that

rkσ¯ < R . (45) k

Set C := { t ∈ (0, φ∞ ) | Γ (t) ∩ [∪k Bkσ¯ ] = ∅ } . The definitions imply that C ⊂ ∪k φ(Bkσ¯ ), and, as a consequence,

|C| ≤ 2∇φ∞ rkσ¯ < 2∇φ∞ R = |Dd
| . k

Hence Dd
\ C = ∅. 3. Let t0 ∈ Dd
\ C. The definition of Dd
implies that |dg(u; Γ (t0 ))| = |deg(u; Γ (t0 )| ≥ d. On the other hand, the definition of C implies that Γ (t0 )∩ (∪k Bkσ¯ ) = ∅, so (40) and the additivity of the degree yield

Variational and Dynamic Problems for the Ginzburg-Landau Functional



d ≤ |dg(u; Γ (t0 ))| ≤

201

|dσk¯ |

¯ ⊂Ω(t ) } {k : Bkσ 0



≤ {k :

¯ ∩∂U =∅ Bkσ

|dσk¯ | }



≤

¯ ∩∂U =∅ } {k : Bkσ

rkσ¯ , σ ¯

by (42). On the other hand, by (45), d=

R rkσ¯ > . σ ¯ σ ¯ k

Therefore we conclude that " (45) is false. 4. By the previous step k rkσ¯ ≥ R = d¯ σ . Hence, by (41),



E
(u) dx ≥ E
(u) dx U

k

≥



¯ U ∩Bkσ

rkσ¯

k

Λ
(¯ σ) σ ¯

σ) . ≥ d Λ
(¯ 2

3 Jacobian and the GL Energy In this section, we show that the Jacobian is bounded by the GL energy. This estimate is the crucial step in a Gamma convergence result. Results of this section are taken from [13]. 3.1 Jacobian estimate The chief result of this section is the following estimate of the Jacobian in terms of the Ginzburg Landau energy. This estimate will be the main ingredient in the compactness result. We give a more precise version of the estimate at the end of the section. We use the notation introduced in the previous section and set
1 E
(u) dx. µ
(V ) := µ
u (V ) = ln(1/) V Theorem 3.1 (Jerrard & Soner [13]). Suppose φ ∈ Cc0,1 (U ) and u ∈ H 1 (U ; IR2 ). For any λ ∈ (1, 2], and  ∈ (0, 1],

202

Halil Mete Soner


    φ Ju dx ≤ πdλ φ∞ + φC 0,1 h
(φ, u, λ),  

(46)

U

3

4 λ
µ (spt(φ)) , dλ = π u

where

(47)

!x" denotes the greatest integer less than or equal to x, and h
(φ, u, λ) ≤ Cα(λ) (1 + µ
u (spt(φ)))(1 + Leb2 (spt(φ))), where α(λ) =

λ−1 12λ

(48)

and C is a constant independent of u, φ, , λ and U .

Note that h
depends on φ only through the support of φ, and on u only through its (linear) dependence on µ
u (spt(φ)). It suffices to consider nonnegative test functions, since we can decompose an arbitrary function φ into its positive and negative parts. So we will assume that φ ≥ 0. By an approximation argument, we may also assume that u is smooth. The main idea behind the above estimate is the following identity, which relies on the co-area formula, integration by parts, and the identity Ju = ∇ × j(u)/2


1 T φ Ju dx = j(u) · t dH1 dt , (49) 2 0 ∂Ω(t) U where, as before, Ω(t) = { x ∈ U | φ(x) > t } , t = unit tangent to ∂Ω(t) =

∇×φ . |∇ × φ|

The proof shows that
j(u) · t dH1 ≈ 2π deg(u; ∂Ω(t)), ∂Ω(t)

for most values of t. The other main point is then to prove that the set of t such that deg(u; ∂Ω(t)) > dλ has Leb1 measure that can be controlled by µ
(spt(φ)). This last point is similar in spirit to results established in [7, 12, 25] for example. We use the lower bounds obtained in the previous section to achieve this. We start the proof of Theorem 3.1 with two simple estimates. Lemma 3.1. For any set A, 


 |A|  1 j(u) · t dH  dt ≤ E
(u) dx .   2 A  ∂Ω(t) spt(φ)

(50)

Variational and Dynamic Problems for the Ginzburg-Landau Functional

For any nonnegative function f ,
T

1 f (x) dH dt ≤ ∇φ∞ 0

∂Ω(t)

f (x) dx .

203

(51)

spt(φ)

Proof. For any t ∈ Reg(φ), Stokes’ Theorem yields

1 j(u) · t dH1 = Ju dx . 2 Ω(t) ∂Ω(t) Since |Ju| ≤ 12 |∇u|2 ≤ E
(u), (50) follows from the above identity. For (51), we calculate, by using the coarea formula,

T
f dH1 dt = f |∇φ| dx ∂Ω(t) spt(φ) 0
≤ ∇φ∞ f dx. spt(φ)

2 We are now in a position to prove Theorem 3.1. In the proof we repeatedly absorb logarithmic factors by using the fact that, if β < α, then α ln(1/) ≤ Cβ , for some C = C(α, β) independent of  ∈ (0, 1]. Proof of Theorem 3.1. 1. Recall that we are writing T = φ∞ . Fix λ ∈ (1, 2] and define dλ := ! πλ µ
(spt(φ))". We define sets A, B ⊂ [0, T ] by B := {t ∈ Reg(φ) : |deg(u; Γ (t))| ≥ dλ + 1 or H1 (γ(t)) ≥ },

(52)

A = Reg(φ) \ B.

(53)

Because almost every t belongs to A ∪ B = Reg(φ), (49) implies that


1 φ Ju dx = j(u) · t dH1 dt 2 A Γ (t) U



1 1 + j(u) · t dH1 dt + j(u) · t dH1 dt 2 A γ(t) 2 B ∂Ω(t) = IA,Γ + IA,γ + IB .

(54)

2. Estimate of IA,Γ Suppose t ∈ A. On Γ (t), |u| ≥ 1/2 by the definition (30), and we set v := u/|u|, so that j(v) = j(u)/|u|2 , and
j(v) · t dH1 = 2π deg(u; Γ (t)). Γ (t)

204

Then

Halil Mete Soner





j(u) · t dH1 = 2π deg(u; Γ (t)) +

Γ (t)

j(u) Γ (t)

|u|2 − 1 · t dH1 . |u|2

Since |j(u) ≤ |u| |∇u|, and since |u| ≥ 1/2 on Γ (t), Cauchy’s inequality and (51) imply that    2



  |u| −1    1  dH1 j(u) · t dH −2πdeg(u; Γ (t))  dt ≤ |∇u|      |u| A Γ (t) A Γ (t)

≤ 4 E
(u) dH1 A

Γ (t)

≤ 4 ln(1/) ∇φ∞ µ
(spt(φ)) . (55) Clearly A ⊂ [0, T ] has measure less than T = φ∞ . Also, by definition of A, if t ∈ A and Γ (t) is nonempty, then |deg(u; Γ (t))| ≤ dλ . It follows that |IA,Γ | ≤ πφ∞ dλ + C1/2 ∇φ∞ µ
(spt(φ)).

(56)

3. Estimate of IA,γ Using Cauchy’s inequality and the elementary fact that x ≤ 1b (1 − x)2 + (1 + 4b ) for all x ∈ IR and b > 0, we have     a 1 b a 1 (1−|u|2 )2 |j(u)| ≤ |u||∇u| ≤ |∇u|2 + 2 |u|2 ≤ |∇u|2 + + (1+ ) , 2 a 2 a2 b 2a 4 for every a, b > 0. We select a = α for α ∈ (0, 1) and b = 2−2α to find |j(u)| ≤ Cα E
(u) + C−α .

(57)

The definition (53) of A implies that |A| ≤ T = φ∞ and that H1 (γ(t)) <  for every t ∈ A, so we can take α = 1/2 and use (51) to find



√
C 1 √ dH1 (dx)dt |IA,γ | ≤ C E (u)dH (x)dt + C  A γ(t) A γ(t) √
√ ≤ C ln(1/)  µ (spt(φ))∇φ∞ + C φ∞ . 4. Estimate of IB To estimate IB we prove that B has small measure. Toward this end we define B1 := {t ∈ Reg(φ) : H1 (γ(t)) ≥ }, B2 := {t ∈ Reg(φ) : Γ (t) is nonempty, and |deg(u; Γ (t))| ≥ dλ + 1}. (58) The estimate of B2 is deferred to the end of this subsection, where we prove

Variational and Dynamic Problems for the Ginzburg-Landau Functional

205

Proposition 3.1. For every λ ∈ (1, 2],  ∈ (0, 1], smooth u : U → IR2 , and nonnegative test function φ ∈ Cc0,1 (U ), 1

1

|B2 | ≤ C1− λ ∇φ∞ (dλ + 1) ≤ C1− λ ∇φ∞ (1 + µ
(spt(φ)).

(59)

For the time being we assume this fact and use it to complete the proof of the theorem. The measure of B1 is easily estimated: using (51) and Lemma 2.3,

1 |B1 | ≤ E
(u)dH1 dt 25 t∈B1 ∂Ω(t) 1 ≤ ∇φ∞ ln( )µ
(spt(φ)). (60)  Clearly |B| ≤ |B1 | + |B2 |, so, by combining (60) and (59), we obtain |B| ≤ C

λ−1 2λ

∇φ∞ (1 + µ
(spt(φ))).

(61)

Finally, we use (50) to estimate |IB | ≤ C

λ−1 3λ

∇φ∞ (1 + µ
(spt(φ)))µ
(spt(φ)).

(62)

5. The previous three steps imply that  
   φJu dx ≤ dλ φ∞ + φC 1 h
0 (φ, u, λ),   for  h
0 (φ, u, λ) ≤ C4α(λ) 1 + µ
(spt(φ)) + (µ
(spt(φ)))2 ,

α(λ) =

λ−1 . 12λ

To complete the proof of the Theorem, note that by (51) and (57) (with α = 2α(λ)) 

T
   φJu dx ≤ |j(u)|dH1 dt   ∂Ω(t) 0
≤ C∇φ∞ 2α(λ) E
(u) + −2α(λ) dx spt(φ)

≤ CφC 1 h
1 (φ, u, λ), for h
1 = α(λ) µ
(spt(φ)) + −2α(λ) Leb2 (spt(φ)). We define h
(φ, u, λ) := min{h
0 , h
1 }, so that (46) clearly holds. It thus suffices to verify that (48) holds, that is, h
(φ, u, λ) = min{h
0 , h
1 } ≤ Cα(λ) (1 + µ
(spt(φ))(1 + Leb2 (spt(φ))),

206

Halil Mete Soner

for some appropriately large constant C. This follows immediately from the definition of h
0 if µ
(spt(φ)) ≤ −3α(λ) , and if not, it follows directly from the definition of h
1 . 2 Note that the result we have proved is in fact somewhat sharper than Theorem 3.1 as stated, in that it not only provides an upper bound for  φJu, but in fact gives an approximate value for the integral. The following corollary states a small technical modification of this sharper estimate. Corollary 3.1. Let U be a bounded, open subset of IR2 , and suppose that φ ∈ Cc0,1 (U ) and u ∈ H 1 (U ; IR2 ). Define Reg(φ), Γ (t) and γ(t) as in (29), (30) and (31) respectively. Then for any λ ∈ (1, 2] and  ∈ (0, 1], there exists a set A = A(φ, u, λ, ) ⊂ (0, φ∞ ) such that |A| ≥ φ∞ − Cα(λ) ∇φ∞ (1 + µ
(spt(φ))); ∀t ∈ A; and

(64)

  deg(u; Γ (t)) dt ≤ φC 1 h
(φ, u, λ),

(65)

Γ (t) is nonempty, and |deg(u; Γ (t))| ≤ dλ 

  φJu − π 

t∈A

(63)

where h
is defined in (48) and dλ is defined in (47). Proof. We cannot take A to be the set defined in (53), as we have now imposed the additional condition that Γ (t) = ∅ for t ∈ A. So we let A˜ be the set formerly known as A, defined in (53), and we define A = {t ∈ A˜ : Γ (t) is nonempty}. ˜ and (65) follows from (55). We Then (64) follows from the definition of A, ˜ claim moreover that A \ A has measure at most ∇φ∞ . In view of (61) and (53), this will suffice to establish (63), and thus to complete the proof of the Corollary. ˜ H1 (γ(t)) < . If To prove our claim, note first that for every t ∈ A, 1 −1 ˜ t ∈ A \ A, then Γ (t) is empty, and so H (φ (t)) = H1 (γ(t)) <  for all t ∈ A˜ \ A. On the other hand, let x0 ∈ U be a point such that φ(x0 ) = φ∞ . If |y − x0 | ≤  then φ(y) ≥ φ∞ −  ∇φ∞ . It follows that B
(x0 ) ⊂ Ω(t) for all t < φ∞ −  ∇φ∞ . Thus the isoperimetric inequality implies that H1 (φ−1 )(t) ≥ 2π. We conclude that if t ∈ A˜ \ A, then t ≥ φ∞ − ∇φ∞ , which proves the claim. 2 We now use Theorem 2.2 and the facts about Λ
to give the proof of Proposition 3.1. Recall that for Proposition 3.1 we want to estimate the measure of a set B2 ⊂ Reg(φ), and from the definition (58) of B2 we see that

Variational and Dynamic Problems for the Ginzburg-Landau Functional

207

λ λ for d∗λ := ! µ
(spt(φ))"+1 ≥ µ
(spt(φ)). (66) π π Proof of Proposition 3.1. We need to show that Dd
∗λ = B2 ∩ {t : t ≥ t
},

1

|B2 | ≤ C1− λ ∇φ∞ d∗λ , Let R := that

|Dd ∗ |

2∇φ∞ . λ

d∗λ := dλ + 1.

From (66) and the definition (32) of t
it suffices to show 1 R ≤ C1− λ . d∗λ

We may assume that dR∗ ≥ , as otherwise the conclusion is obvious. Then λ (22), Theorem 2.2, and the choice (66) of d∗λ imply that        R R 1 1 ln π ln = − π ln d∗λ π d∗λ      R 1 1 + C − ln ≤ Λ
π d∗λ      1 1 1
µ ln (spt(φ)) + C − ln ≤ πd∗λ       1 1 − 1 ln + C. ≤ λ  2 3.2 Compactness in two dimensions In this section we consider a sequence of functions u
∈ H 1 (U ; IR2 ), where U is a bounded open subset of IR2 and the renormalized Ginzburg-Landau energy is uniformly bounded: KU := sup µ
(U ) < ∞ ,

µ
:= µ
u .

(67)


∈(0,1]

We will show that under this assumption, the Jacobian is compact in the dual norm (C 0,β )∗ for every β ∈ (0, 1]. We refer to §5 and to [13] for a compactness in higher dimensions. We introduce the Jacobian (signed) measure

Ju (E) := det (∇u
) dx , E ⊂ U. E

Since det (∇u
) = 12 ∇ × j(u
) for j(u
) := u
× ∇u
,

208

Halil Mete Soner


IR2

φ dJu
=

1 2


IR2

∇ × φ(x) · j(u
)(x) dx ,

∀φ ∈ Cc1 (U ),

where for a scalar function φ, we write ∇ × φ := (φx2 , −φx1 ). Theorem 3.2. Let {u
} ⊂ H 1 (U ; IR2 ) satisfy (67). Then there exists a subsequence n converging to zero and Radon measure J such that Ju
n ∗  a0,βsigned converges to J in the dual norm Cc for every β ∈ (0, 1]. Moreover, there ⊂ U and integers k such that are {ai }N i i=1 J =π

N

ki δai ,

and

|J|(U ) = π

i=1



|ki | ≤ KU .

i

Finally, if µ
converges weakly to a limit µ, then J  µ, and µ almost every x.

dJ dµ (x)

≤ 1 for

We will first prove Proposition 3.2. Assume (67). Then, Ju
can be written in the form Ju
= J0
+ J1
where J0
and J1
are signed measures such that J0
(C 0 )∗ ≤ C, and

J1
(Cc0,1 )∗ ≤ Cα

(68)

for some α > 0 and a constant C depending only on the constant KU in (67). Proof. 1. In light of the assumption µ
(U ) ≤ K, Theorem 3.1 (with λ = 2 and α = 1/24, for example) implies that
φJu
≤ Cφ∞ + Cα ∇φ∞ for all φ ∈ Cc0,1 (U ). (69) We write δ = α , and we define Uδ = {x ∈ U : dist(x, ∂U ) > δ}. Let  1 if x ∈ U2δ χδ = 0 if not. We define J0
:= χδ (η δ ∗Ju
), where η δ is a standard mollifier with support in Bδ (0). We then define J1
:= Ju
− J0
. Suppose that φ is a C 1 test function vanishing on ∂U , and note that


φ J0 dx = η δ ∗ (χδ φ)Ju
dx.

Variational and Dynamic Problems for the Ginzburg-Landau Functional

209

We write φδ := η δ ∗ (χδ φ). It is clear that φδ is compactly supported in U , and one easily checks that φδ ∞ ≤ χδ φ∞ ≤ φ∞ , Then (69) implies that

∇φδ ∞ ≤

C C χδ φ∞ ≤ φ∞ . δ δ


φ J0
dx ≤ Cφ∞ .

2. We now estimate J1
. Given φ ∈ C01 (U ), write φ1 := min{φ, 2δ∇φ∞ },

φ2 := φ − φ1 .

It is clear that φ ≤ 2δ∇φ∞ in U \ U2δ , so φ2 is supported in U2δ . From the definitions,

φ1 J1
dx = (φ1 − η δ ∗ (χδ φ1 ))Ju
dx. It is clear that φ1 ∞ ≤ 2δ∇φ∞ ,

∇φ1 ∞ ≤ ∇φ∞ .

Similarly, η δ ∗ (χδ φ1 ) satisfies η δ ∗ (χδ φ1 )∞ ≤ 2δ∇φ∞ ,

∇η δ ∗ (χδ φ1 )∞ ≤

C φ1 ∞ ≤ C∇φ∞ . δ

So (69) implies that
φ1 J1
dx ≤ Cδ∇φ∞ = Cα ∇φ∞ . Finally, since φ2 is supported in U2δ ,



δ
φ2 J1 dx = (φ2 − η ∗ (χδ φ2 ))Ju dx = (φ2 − η δ ∗ φ2 )Ju
dx. It is easy to check that φ2 − η δ ∗ φ2 ∞ ≤ Cδ∇φ∞ ,

∇(φ2 − η δ ∗ φ2 )∞ ≤ C∇φ∞ .

So we again use (69) to conclude
φ2 J1
dx ≤ Cδ∇φ∞ = Cα ∇φ∞ . 2 Once we have the above decomposition, the compactness of the sequence Ju
follows from soft arguments.

210

Halil Mete Soner

Lemma 3.2. If ν is a Radon measure on U , then ν(Cc0,α )∗ ≤ Cνα ν1−α (C 0 )∗ . (C 0,1 )∗ c

(70)

c

Proof. Since U is bounded and we are considering compactly supported functions, the H¨ older seminorm is in fact a norm and is topologically equivalent to the usual C 0,α norm. So for this lemma we set φCc0,α (U ) := [u]C 0,α = sup x =y

|φ(x) − φ(y)| , |x − y|α

α ∈ (0, 1].

Fix φ ∈ Cc0,α , and let φ˜
= η
∗ φ, where η
is a smoothing kernel and  will be chosen later. Then one easily checks that φ˜
C 0,1 ≤ Cα−1 φC 0,α := M
,

φ − φ˜
C 0 ≤ Cα φC 0,α .

(71)

In particular, |φ˜
| ≤ Cα φC 0,α on ∂U . We next modify φ˜
so that it vanishes on ∂U while continuing to satisfy the above estimates. Let   + − v(x) = sup φ˜
(y) + M
|x − y| . u(x) = sup φ˜
(y) − M
|x − y| , y∈∂U

y∈∂U

Then one easily checks that φ˜
= u − v on ∂U . Moreover, if we define φ
:= φ˜
− u + v, then φ
satisfies the estimates in (71) and also vanishes on ∂U . So we have



φdν = φ dµ + (φ − φ
) dν ≤ φ
C 0,1 ν(Cc0,1 )∗ + φ − φ
C 0 ν(C 0 )∗   ≤ CφC 0,α α−1 ν(Cc0,1 )∗ + α ν(C 0 )∗ . Taking  = ν(Cc0,1 )∗ /ν(C 0 )∗ gives the conclusion of the lemma.

2

Lemma 3.3. If α > 0, then (C 0 )∗ ⊂⊂ (C 0,α )∗ . Proof. The Arzela-Ascoli Theorem implies that any sequence that is bounded on C 0,α is precompact in C 0 . The lemma follows by duality. More concretely: given a sequence of measures bounded in (C 0 )∗ , we can extract a subsequence, say µn that converges to a limit µ in the weak-* topology. We must show that this sequence converges in norm in (C 0,α )∗ . If not, then we can find a sequence of functions ψn with ψn C 0,α ≤ 1 such that
(72) ψn d(µn − µ) ≥ c0 > 0, for all n. However, the Arzela-Ascoli theorem implies that, upon passing to a subsequence, ψn converges to some limit ψ uniformly, whence (72) is impossible. 2 We now prove

Variational and Dynamic Problems for the Ginzburg-Landau Functional

211

Theorem 3.3. Assume (67). Then Ju
is strongly precompact in (C 0,β )∗ for all β > 0. Proof. By Proposition 3.2 we can write Ju
= J0
+ J1
, where the two measures on the right-hand side satisfy (68). Fix any β ∈ (0, 1]. Lemma 3.3 implies that {J0
} is precompact in 0,β ∗ (C ) ⊂ (Cc0,β )∗ . Also, it is clear from the definitions that 1 J1
(C 0 )∗ ≤ Ju
L1 + J0
(C 0 )∗ ≤ C∇u
2L2 + C ≤ K ln( ).  So together with (68) and the interpolation inequality (70) this implies that J1
(Cc0,β )∗ → 0 as  → 0. 2 Remark 3.1. The above result is sharp in the sense that Ju
need not be precompact, or even weakly precompact, in (C 0 )∗ . To see this, consider the sequence of functions 1 x y u
(x, y) = (1, 0) + 2 (ln( ))1/2 (cos( 2 ), sin( 2 ))    on the open unit disk D in the plane. One easily verifies that µ
(D) ≤ C, and that Ju
(C 0 )∗ = Ju
L1 ≥ c−1 ln( 1
). In particular, since Ju
(C 0 )∗ is unbounded, the Uniform Boundedness Principle implies that the sequence cannot converge weakly in (C 0 )∗ . Remark 3.2. Suppose ν
is any sequence of measures on a bounded open set U ⊂ IRm , and that
1 φdν
≤ Cφ∞ + Cα ∇φ∞ , |ν
|(U ) ≤ K ln( ),  for some α > 0. The arguments given above then show, with essentially no change, that {ν
} is precompact in (C 0,β )∗ for all β ∈ (0, 1]. We are now in a position to give the Proof of Theorem 3.2. Suppose {u
}
∈(0,1] ⊂ H 1 (U ; IR2 ) is a sequence satisfying (67). By an approximation argument, we may assume that in fact each u
is smooth. In view of Theorem 3.3, we can find a measure J and a subsequence n such that Ju
n → J in (Cc0,β )∗ for every β ∈ (0, 1]. 1. Since µ
n is a sequence of uniformly bounded, nonnegative Radon measures, we may assume upon passing to a further subsequence (still labeled n ) that there is a Radon measure µ such that µn := µ
n

∗

µ,

in the weak∗ topology of Radon measures in U . For x ∈ U , set

212

Halil Mete Soner

Θ(x) := lim µ (Br (x) ∩ U ) . r↓0

We first claim that J is supported only on the points with Θ(x) ≥ π. Indeed, suppose that Θ(x0 ) < π at some x0 ∈ U . Then there exists some r0 > 0 and a number α < π such that µn (Br0 (x0 )) ≤ α < π for all sufficiently large n. Then Theorem 3.1 with λ = a+π 2a > 1 immediately implies that

φ dJ(x) = lim φ Ju
n dx = 0 n→∞

for all smooth φ with support in Br0 (x0 ), since dλ = 0 for such φ. Thus x0 ∈ spt(J). Since µ is bounded on U , there are finitely many points {ai }i ⊂ U such that Θ(ai ) ≥ π . Therefore there are constants ci such that the limit measure J satisfies

J =π ci δai . i

We need to prove that ci ’s are integers and that π|ci | ≤ Θ(ai ) for all i; this will immediately imply all the remaining conclusions of Theorem 3.2. 2. Choose r1 ≤ 1 so that Br1 (a1 ) does not intersect {ai }i>1 ∪ ∂U . We may also assume, taking r1 smaller if necessary, that there exists some λ > 1 and an integer N0 such that λ 1 dλ := ! µn (Br1 (a1 ))" ≤ Θ(a1 ) π π

∀n ≥ N0 .

(73)

We first apply Theorem 3.1 to the function φ(x) := (r1 − |x − a1 |)+ , which is supported in Br1 (a1 ). Let An = A(φ, u
n , λ, n ) be the set whose existence is asserted in Theorem 3.1. Note that if t ∈ An , then Γφ,un (t) is nonempty, which is to say that there is a component of φ−1 (t) on which min |u| ≥ 1/2. However, φ−1 (t) = ∂Br1 −t (a1 ) is connected, so in fact Γφ,un (t) = ∂Br1 −t (a1 ) for all t ∈ An . So for every t ∈ An and n ≥ N0 , Theorem 3.1 and the choice of λ imply that min x∈∂Br1 −t (a1 )

|u
n | ≥

1 , 2

|deg(u
n ; ∂Br1 −t (a1 ))| ≤ dλ ≤

1 Θ(a1 ). π

It follows that for all such n there is an integer d(n) such that the set Snd(n) := {r ∈ [0, r1 ] :

min |u
n | >

∂Br (ai )

1 , deg(u
n ; ∂Br ) = d(n)} 2

Variational and Dynamic Problems for the Ginzburg-Landau Functional

213

d(n)

has measure at least k0 := r1 /(3dλ ). Note also that Sn is open, since u
n is by assumption continuous (indeed, smooth). We can therefore find an open d(n) set Σn ⊂ Sn such that |Σn | = k0 . 3. We now define new test functions ψ n as follows. First let f n (r) = |[r, r1 ] ∩ Σn | . We then define ψ n (x) = f n (|x − a1 |). One can then check that t is a regular value of ψ n if and only if (ψ n )−1 (t) = ∂Br (a1 )

for some r ∈ Σn .

In particular, deg(u; (ψ n )−1 (t)) = d(n) for a.e. 0 < t < ψ n ∞ = k0 . One can then easily check, using Theorem 3.1, that
ψ n Ju
n dx = πd(n)k0 + O(α ). On the other hand, since"the functions ψ n are uniformly bounded in Cc0,1 and since Ju
n → J = π ci δai in Cc0,1 (U )∗ , 

      n
n n n
n    0 = lim  ψ Ju dx − πc1 ψ (a1 ) = lim  ψ Ju dx − πc1 k0  . n

n

Comparing the last two equations, we find that d(n) = c1 for all sufficiently large n. In particular, c1 is an integer and |c1 | ≤ dλ ≤ π1 Θ(a1 ), which is what we needed to show. 2

4 Gamma Limit Suppose that a sequence of functionals Jn on a Banach space X is given. We assume that they are bounded from below and by adding a constant, if necessary, we may assume that these functionals are nonnegative, i.e., Jn : X → [0, ∞]. The Gamma limit, J, of these functional in the topology of X is roughly given by J(x) := lim inf{ Jn (xn ) |(n, xn ) → (∞, x) }. More precisely, the Gamma limit is a nonnegative functional J : X → [0, ∞], satisfying the following two conditions

214

Halil Mete Soner

– Let {xn } ⊂ X be a convergent sequence with limit x ∈ X, i.e., lim xn − xX = 0.

n→∞

Then, lim inf Jn (xn ) ≥ J(x). n→∞

– For any x ∈ X, there exists a sequence {x∗n } satisfying lim x∗n − xX = 0,

n→∞

and

lim Jn (x∗n ) = J(x).

n→∞

Generally an accompanying compactness result is useful. Such a compactness result states that, if for a sequence {xn } ⊂ X sup Jn (xn ) < ∞,

(74)

n

then the set {xn } is compact in X. An immediate corollary to a Gamma limit and to the compactness result is this: Let yn∗ be a minimizer of Jn , and they satisfy (74). Then, by compactness on a subsequence, denoted by n again, yn converges to a point y ∗ ∈ X. By the first condition on the Gamma limit we know that J(y ∗ ) ≤ lim inf Jn (yn ). n

{x∗n }

be the sequence in the second condition Let x be any point in X and let on the Gamma limit. Since yn ’s minimize Jn , J(x) = lim Jn (xn ) n

≥ lim sup Jn (yn ) n

≥ J(y ∗ ). Hence, any limit y ∗ of the minimizing sequence yn is a minimizer of the Gamma limit J. In practice, this method is used to a construct minimizer of a given functional J. Given J, we construct a “regular” functionals Jn such that the Gamma limit of this sequence is J. Then, we obtain minimizers of Jn by standard methods and then apply the above method to construct a minimizer of J. The sequence Jn is often called a relaxation of J. We refer to the book of Dal Maso [8] for more information. In this section we calculate the Gamma limit of
1 E
(u) dx. I
(u) := ln(1/) First we introduce a function space which is needed in order to state the Gamma limit.

Variational and Dynamic Problems for the Ginzburg-Landau Functional

215

4.1 Functions of BnV Motivated by the analysis of I
and the central role of BV in the scalar case, Jerrard and the author introduced and studied a class of functions called BnV in [14]; a short summary is provided in [15]. It turns out that the Gamma limit of the functional I
is finite only for functions u ∈ B2V . For that reason, we give a brief discussion of BnV . We refer to [14, 15] for more information. Briefly, function u ∈ W 1,n−1 (U ; IRn ), for U ⊂ IRm , m ≥ n, is said to belong to BnV if the weak determinants of all n by n submatrices of the gradient matrix ∇u are signed Radon measures. Here we only we give the precise definition of B2V and refer to [14] for higher dimensional case. For U ⊂ IRm → IR2 with m ≥ 2 we view the Jacobian as a measure taking values in the exterior algebra Λ2 IRm . For every n (and in particular for n = 2) we endow Λn IRm with the natural inner product structure, which we denote (·, ·), and for a multivector v ∈ Λn IRm we write |v| = (v, v)1/2 . If u ∈ W 1,1 (U ; IR2 ) we define j(u) =

m

u × uxi dxi ,

(75)

i=1

and if j(u) ∈ L1loc , we define Ju =

1 d j(u) 2

in the sense of distributions,

(76)

1 where d is the exterior derivative. Thus if u ∈ Hloc , then

Ju =

i<j

J ij u dxi ∧ dxj =

1 ij J u dxi ∧ dxj , 2 i,j

where J ij u = −J ji u = uxi × uxj = det(uxi , uxj ). For sufficiently differentiable u : IRm → IRn one can define in a similar way Ju as a measure taking values in Λn IRm . We omit the most general definition as we will not need it here. Here we give the precise definition of B2V . Let U ⊂ IRm with m ≥ 2. Definition. We say that u belongs to B2V (U ; IR2 ) if both of the conditions are satisfied – j(u) ∈ L1loc (U ; IR2 ), – Ju is a Radon measure with values in Λ2 IRm . A priori Ju is only a distribution; we say that u ∈ B2V if it happens to be a measure. Also there are several conditions that ensure that j(u) ∈ L1 . For instance if u ∈ W 1,1 ∩ L∞ j(u) ∈ L1 .

216

Halil Mete Soner

The class BnV is very closely related to the Cartesian Currents of Giaquinta, Modica and Soucek [10, 11]. This connection is discussed in detail in [14]. In [14] it is shown that if u ∈ B2V (IRm ; S 1 ), then the Jacobian measures Ju is supported on an m − 2 dimensional rectifiable set. In particular, if u ∈ B2V (U ; S 1 ) and U ⊂ IR2 , then there are {ai } ⊂ U and integers ki such that

ki δai . Ju = π i

This is interpreted as encoding the location and degree of the topological singularities of u. Here we outline only the proof in the case of u ∈ B2V (U ; S 1 ) with U ⊂ 2 IR , and refer to [14] for the higher dimensional result. Lemma 4.1. Let U ⊂ IR2 and u ∈ B2V (U ; IR2 ). Then

Ju = π kj δaj ,

(77)

j

for finite collections of points {aj } ⊂ U and integers kj . Outline of Proof. In this case, Ju is a scalar valued, signed Radon measure. Hence for every x ∈ U , the following limit exists d(x) := lim dr (x), r↓0

dr (x) := Ju(Br (x)).

We claim that for every x ∈ U , and for almost every r > 0, dr (x)/π is an integer. Indeed, since Ju = ∇×j(u), for any smooth function φ ∈ Cc∞ (U ; IR1 ), by integration by parts

φJu(dx) = ∇φ · j(u) dx. U

U

Formally if we take φ to be the characteristic function of the set Br (x), we obtain
Ju(dx) Ju(Br (x)) = Br (x))
1 = j(u) · t dH1 . 2 ∂Br (x) By approximation, we may show that the above identity holds if u is sufficiently smooth. For u ∈ B2V , in [14], the above identity is proved for almost every r > 0. Moreover, since u ∈ W 1,1 (U ; S 1 ), for almost every r > 0, u ∈ W 1,1 (Br (x); S 1 ). Hence, for these values of r, u is absolutely continuous with values in S 1 . Then, by the degree formulae discussed earlier,

Variational and Dynamic Problems for the Ginzburg-Landau Functional

1 2

217


j(u) · t dH1 = π deg(u; ∂Br (x)). ∂Br (x)

Therefore dr (x)/π is an integer for almost every r. Since the limit of dr (x) exists, we conclude that for every x ∈ there exists r(x) > 0 such that dr (x) = d(x),

∀ r ∈ (0, r(x)],

and d(x)/π ∈ Z,

∀ x ∈ U.

Since |Ju|(U ) is finite, by a simple covering argument these yield the desired result. 2

4.2 Gamma limit of I
Let U be an open bounded subset of IR2 with a smooth boundary. In this section we study the Γ limit of the functionals
1 1 1 I
(u) := |∇u|2 + 2 (1 − |u|2 )2 dx , ln(1/) U 2 4 as  tends to zero and show that the limiting functional is  " if u ∈ B2V (U ; S 1 ) ,  |Ju|(U ) = π i |ki | , I(u) :=  +∞ , if u ∈ B2V (U ; S 1 ) . Theorem 4.1. The Γ limit of I
in the topology of W 1,1 (U ; IR2 ) is equal to I, i.e., for every sequence u
converging to u in W 1,1 (U ; IR2 ), lim inf I
(u
) ≥ I(u) ,
→0

(78)

and for every u ∈ B2V (U ; S 1 ), there exist functions u
converging to u in W 1,1 (U ; IR2 ) satisfying lim inf I
n (u
) = I(u) .
→0

(79)

A similar result also holds in higher dimensions [13]. The above result is proved independently in [1] and in [13]. Remark 4.1. In view of our compactness result and our introductory discussion of Gamma Limit, the Banach space W 1,1 is not appropriate. Since we do not have a compactness result in W 1,1 , and the only compactness result is for the Jacobian, it is more natural to consider the Gamma Limit in the space of equivalent classes of functions with a topology equivalent to the convergence of the Jacobian. However, here and in [13], we chose to work with W 1,1 as it is a standard Banach space.

218

Halil Mete Soner

Proof. We start with the proof of (78). Suppose that u
converges to u in W 1,1 (U ; IR2 ). We assume that lim inf I
(u
) < ∞ ,


as there would be nothing to prove otherwise. 1. By the Compactness Theorem 3.2, there exists a subsequence n converging to zero such that the Jacobian measure Ju
n converges to a Radon measure J in (Cc0,β )∗ for all β > 0. We claim that J = Ju. In particular this will show that u ∈ B2V (U ; S 1 ). To simplify the notation, set un := u
n . 2. We directly estimate that     2     j(un ) − j(un )  ≤ |un ||∇un |  |un | ∧ 1 − 1     |un |2 ∧ 1 |un |2 ∧ 1  |1 − |un |2 | χ|un |≥1 = |∇un | |un |   1 1 2 2 2 ≤ n |∇un | + 2 (1 − |un | ) . 2 2n Hence,


  j(un )   lim dx = 0 . j(un ) − n→∞ U  |un |2 ∧ 1 

3. Set vn := un /(|un |2 ∧ 1) so that 1 j(un ) − j(u) = vn × ∇un − u × ∇u |un |2 ∧ 1 = vn × (∇un − ∇u) + (vn − u) × ∇u . Hence,     1    |un |2 ∧ 1 j(un ) − j(u) ≤ |vn ||∇un − ∇u| + |vn − u||∇u| ≤ |∇un − ∇u| + |vn − u||∇u| . Since un converges to u in W 1,1 (U ; IR2 ), there exists a subsequence, denoted by n again, so that un converges to u almost everywhere. Hence |vn − u||∇u| converges to zero almost everywhere and also it is less than 2|∇u|. So we may use the dominated convergence theorem to conclude that 
   1   dx = 0 . j(u ) − j(u) lim n   2 n→∞ |un | ∧ 1 U 4. Steps 2 and 3 imply that on a subsequence j(un ) converges to j(u) in L1 . Hence, Ju
n converges to Ju in the sense of distributions. This implies

Variational and Dynamic Problems for the Ginzburg-Landau Functional

219

that J = Ju. Since by Theorem 3.2, J is a Radon measure, so is Ju and therefore u ∈ B2V (U ; IR2 ). It is also clear that |u| = 1 almost everywhere. Hence, u ∈ B2V (U ; S 1 ). 5. The Jacobian estimate (46) implies that 

         φ Ju(dx)  = lim  φ Jun (dx)   n→∞

U

U

≤ λφ∞ lim inf I
n (un ) , n→∞

for every λ > 1. Hence, 
 lim inf I (un ) ≥ sup {  n→∞

  φ Ju(dx)  : φ∞ ≤ 1 }


n

U

= |Ju|(U ) = I(u) . This proves (78). 6. We continue by proving the Γ -limit upper bound (79). Fix u ∈ B2V (U ; S 1 ). As remarked above, it is shown in [14] that Ju must have the form

Ju = π kj δaj . j

It suffices to show that, given any sufficiently small δ > 0, there exists a sequence of functions {v
} ⊂ H 1 (U ; IR2 ) such that

|kj |, lim sup v
− uW 1,1 (U ) ≤ Cδ. I
(v
) → π


To do this, fix some small δ > 0. Let r0 > 0 be a number such that the balls {B2r (aj )} are pairwise disjoint and do not intersect ∂U , whenever r ≤ r0 , and select some r > 0 such that


|∇u| dx ≤ δ, r ≤ min{r0 , δ}. (80) j

B2r (aj )

For any s > 0, let Us denote U \ ∪j Bs (aj ). Demengel [9] proves that if V is an open subset of IR2 , then smooth functions taking values in S 1 are dense in the subspace {w ∈ W 1,1 (V ; S 1 ) : Jw = 0}. Since Ju = 0 on Ur , this implies that there exists a function v ∈ C ∞ (Ur , S 1 ) such that u − vW 1,1 (Ur ) ≤ δ.

(81)

Demengel’s proof in fact shows that we may also assume that j(u) − j(v)L1 (Ur ) ≤ δ.

(82)

220

Halil Mete Soner

7. Clearly (80) and (81) imply that


j

r

2r


|∇v(x)| dH1 (x)ds = ∂Bs (aj )


j

B2r \Br (aj )

|∇v| dx ≤ 2δ.

So, for each j, we can find some number rj ∈ [r, 2r] such that
2δ . |∇v(x)| dH1 (x) ≤ r ∂Brj (aj )

(83)

We also claim that deg(v; ∂Brj (aj )) = kj ,

(84) 1

if δ is sufficiently small. Indeed, since v is smooth and S -valued it is clear that s → deg(v; ∂Bs (aj )) is constant for s ∈ [r, 2r0 ], so we only need to verify that this constant must equal kj . To do this, note that if φ is any function of ¯ − aj |) that is constant on Br (aj ) and has its support the form φ(x) = φ(|x in B2r0 (aj ), then

1 1 ∞ ∇ × φ · j(v) dx = deg(v; φ−1 (s)) ds = πφ(aj )deg(v; ∂Brj (aj )) 2 0 2 and

1 2





∇ × φ · j(u) dx =

φ dJu = πφ(aj )kj .

If δ is small enough, (84) follows from these two identities and (82), since ∇ × φ is supported in Ur . 8. We claim that for each j there exists smooth functions vj
, defined in Brj (aj ) such that vj
(x) = v(x) for x ∈ ∂Brj (aj ),

1
|∇vj |dx ≤ Cδ, and lim E
(vj
)dx = π|kj |.
→0 | ln | Brj (aj ) Brj (aj ) (85) To see this, fix some j. We may assume without loss of generality that aj = 0, and due to (84) we can write v(x) = exp[i(kj θ + αj + ψ(x))]

for x ∈ ∂Brj ,

where αj is a constant, ψ is a smooth, single-valued function on ∂Brj , and θ x as usual satisfies |x| = (cos θ, sin θ). We are identifying IR2 ∼ I in the usual =C way. We extend ψ to be homogeneous of degree zero on IR2 \ {0}, and we define

Variational and Dynamic Problems for the Ginzburg-Landau Functional

vj
(x) = exp[i(kj θ + αj +

2|x| − rj ψ(x))] rj

221

1 rj ≤ |x| ≤ rj . 2

if

For |x| ≤ 12 rj we define vj
(x) to be a minimizer of
E
(w) dx, Brj /2

subject to the boundary conditions w = exp[i(kj θ + αj )] on ∂Brj /2 . Since vj
restricted to the annulus Brj \Brj /2 is just a fixed smooth function of unit modulus, independent of , it is clear that

1 1 1

lim |∇vj
|2 dx = 0. E (vj ) dx = lim
| ln | B \B
| ln | B \B 2 rj rj r /2 r /2 j

j

Also, using (83) one can check that
Brj \Brj /2 (aj )

|∇vj
|dx ≤ Cδ.

Finally, the book of Bethuel, Brezis, and H´elein gives a detailed description of the asymptotics of Ginzburg-Landau energy-minimizers, and their results imply that

1 lim E
(vj
) dx = π|kj |, lim sup |∇vj
| dx ≤ Crj ≤ Cδ.
| ln | B
B r /2 r /2 j

j

Putting these facts together we find that the sequence {vj
} has the properties specified in (85). 9. Finally we define   v(x) if x ∈ U \ ∪j Brj (aj ) v
(x) = . vj
(x) if x ∈ Brj (aj ) Since v is a fixed smooth function and |v| ≡ 1, | ln1
| E
(v) = | ln1
| |∇v|2 tends to zero uniformly as  → 0. Thus it is clear from (85) that





1 1

E (v )dx = lim E
(vj
)dx = π |kj |. lim
→0 | ln | U
→0 | ln | B (a ) rj j j j Also, u − v
W 1,1 (U ) ≤ u − vW 1,1 (Ur )

+ (uW 1,1 (Brj (aj )) + vj
W 1,1 (Brj (aj )) ) j

≤ Cδ, by (80), (81), and (85). So the sequence {v
} has all the required properties. 2

222

Halil Mete Soner

5 Compactness in Higher Dimensions Now suppose U is a bounded, open subset of IRm with m ≥ 3. In this section we will show that if {u
}
∈(0,1] ⊂ H 1 (U ; IR2 ) is a sequence of functions such that the normalized Ginzburg-Landau energy measure µ
(U ) is uniformly bounded, then the Jacobians Ju
are precompact in (C 0,β )∗ for all β > 0, and any limit is rectifiable. In addition, we prove that ¯ ) ≤ lim inf µ
(U ). |J|(U This is not a full Γ -convergence result, but it shows that the mass of the Jacobian is a reasonable candidate for the Γ -limit. We also believe that the compactness result and the upper bound for the Jacobian (ie, lower bound for the energy) are interesting and will be useful in other contexts. We start by defining some of the terms used above. We remark that good general references for this material include Giaquinta et. al [10] and Simon [28]. Let Λ2 IRm , j(u) and Ju be as in subsection 4.1. A set M ⊂ IRm is said to be a k-dimensional rectifiable set if there are Lipschitz functions fj : IRk → IRm and measurable subsets Aj of IRk such that  M = M0 ∪ ∪∞ Hk (M0 ) = 0. j=1 fj (Aj ) , Thus, in a precise measure theoretic sense, a k-dimensional rectifiable set is not much worse than a k-dimensional Lipschitz submanifold. Rectifiable sets can also be characterized by the fact that they have k-dimensional approximate tangent spaces Hk almost everywhere; see [28] or [10]. Suppose that M is an oriented, rectifiable (m − n)-dimensional subset of IRm , and for Hm−n almost every x ∈ M , let ν(x) ∈ Λn IRm be the unit nvector representing the appropriately oriented normal space to M . (It is more convenient for our purposes to work with normal spaces rather than tangent spaces.) Suppose also that θ : M → IN is a Hm−n -integrable function. One can define a measure J taking values in Λn IRm by

φ(x)J(dx) = φ(x) · ν(x)θ(x)Hm−n (dx) ∀φ ∈ C 0 (IRm ; Λn IRm ). M

(86) We say that a measure J taking values in Λn IRm is (m − n)-dimensional integer multiplicity rectifiable (or more briefly, integer multiplicity rectifiable) if it has the form (86) for some rectifiable set M and an integer-valued function θ as above. The class of functions for which Ju is a measure is denoted BnV (U, IRn ) and was defined and studied in [14]. In particular we prove there that if u ∈ BnV (IRm , S n−1 ) then Ju/ωn is integer multiplicity rectifiable, where ωn is the volume of the unit ball in IRn . This is deduced as a consequence of a more general rectifiability criterion which we recall here.

Variational and Dynamic Problems for the Ginzburg-Landau Functional

223

Let J be a measure on a subset U ⊂ IRm taking values in Λn IRm , where n ≤ m. We can write J in the form J = ν|J|, where |J| is a nonnegative Radon measure, and ν is a |J|-measurable function taking values in Λn IRm such that |ν(x)| = 1 at |J|-a.e. x ∈ U . Suppose that e1 , ..., em is an orthonormal basis for IRm . Given any point x ∈ IRm , we write yi = x · ei if i = 1, ..., m − n; and zi = x · em−n+i if i ∈ 1, . . . , n. We write IRym−n to denote the span of {ei }m−n i=1 . Similarly, m . Thus we identify points x ∈ I R with corresponding IRzn = span{ei }m i=m−n+1 (y, z) ∈ IRym−n × IRzn . Let dz := dz 1 ∧ . . . ∧ dz n , and let J z denote the scalar signed measure defined by J z := (dz, ν)|J|. We say that J z is locally represented by slices Jy (dz) if, given any open set O ⊂ U of the form O = Oy × Oz , with Oy ⊂ IRym−n and Oz ⊂ IRzn , there exist signed Radon measures Jy (dz) on Oz for a.e. y ∈ Oy , such that


z φJ = φ(y, z)Jy (dz) dy (87) Oy

Oz

for all continuous φ with compact support in O. We say that a statement holds for a.e. Jy (dz) if, for every open set O as above, it is valid for a.e. y ∈ Oy . In [14] we prove the following Theorem 5.1. Suppose that J is a Radon measure on U ⊂ IRm taking values in Λn IRm , and also that dJ = 0 in the sense of distributions. Suppose also that for every choice on an orthonormal basis {ei }m i=1 (determining a decomposition of IRm into IRym−n × IRzn ) J z is represented locally by slices, and that for a.e. y ∈ Oy these slices have the form Jy (dz) =

K

di δai (dz),

i=1

for an integers K and di , and points ai ∈ Oz . Then J is rectifiable. A much more general version of this result was later established by Ambrosio and Kirchheim [2]. A similar theorem in somewhat different and very general setting was proved independently (and slightly earlier) by White [33]. We will need Theorem 5.1 to prove Theorem 5.2 (Jerrard & Soner [13]). Let U ⊂ IRm , and suppose that {u
}
∈(0,1] is a collection of functions in W 1,2 (U ; IR2 ) such that µ
(U ) ≤ KU < ∞ for all . Then there exists a subsequence n → 0 such that (i) Ju
n converges to a limit J¯ in the (C 0,α )∗ norm for every α > 0;

224

Halil Mete Soner

(ii) For any choice of basis {ei } for IRm (determining a decomposition of IRm into IRym−2 × IRz2 ), J¯z is represented locally by slices J¯y (dz), and for a.e. "K y these slices have the form J¯y (dz) = π i=1 di δai , with di ∈ ZZ for all i. ¯ (iii) dJ¯ = 0 in the sense of distributions, and J/π is integer multiplicity rectifiable; ¯ µ (iv) Finally, if µ ¯ is any weak limit of a subsequence of µ
n , then |J| ¯, and ¯ d|J| ¯ ≤ 1. In particular, | J|(U ) ≤ K . U d¯ µ Remark 5.1. For any J¯ as above, (iii) and the definition of rectifiability imply that a lower density bound lim inf r→0

¯ r (x)) |J|(B ≥ π, m−2 H (Br (x))

¯ almost every x. Also, if µ for |J| ¯ is as in (iv), then clearly the m − 2¯ dimensional density of µ is greater than m − 2-dimensional density of J. In particular, µ ¯(Br (x)) ≥ π, lim inf m−2 r→0 H (Br (x)) ¯ almost every x. for |J| The basic idea of the proof is to decompose a component of Ju
, for example J m−1,m u
, into two-dimensional slices, say Jy
(dz), and to use the two-dimensional estimates on each slice. Arguing in this fashion, it is quite easy to obtain uniform estimates for J m,m−1 u
in certain weak spaces, and these imply (i) by results of the previous section. To prove (ii), it is convenient to view the sliced measures Jy
(dz) as constituting a function mapping IRym−2 into Cc1 (IRz2 )∗ ; the latter is a space that contains measures on IRz2 and is endowed with a rather weak topology. Claim (i) can be seen as assertion that the function y → Jy
(dz) is precompact in some weak sense. What one would like to do is to show that in fact y → Jy
(dz) is precompact in some stronger sense, for example in L1 (IRym−2 ; (Cc1 (IRz2 )∗ ), so that one can extract a subsequence that converges to some limiting function y → J¯y (dz) in L1 . In particular, after passing to a further subsequence we could then assume that Jy
n (dz) → J¯y (dz) for almost every y. In addition, by our two-dimensional results, for almost every y, one can find a subse
quence nm → 0 (in general depending on y) such that Jynm (dz) converges to some limit that has the form sought in (ii). By combining these results one can hope to show that in fact π1 J¯y (dz) is a sum of point masses with integer multiplicities. The key point is then to establish some sort of strong compactness of the sequence of functions y → Jy
(dz) as  → 0. We do this using the observation from [14, 15] that the total variation of y → Jy
(dz) in the (Cc0,1 )∗ norm can be estimated by controlling “orthogonal” components of Ju
, which is

Variational and Dynamic Problems for the Ginzburg-Landau Functional

225

already done in the proof of (i). Using this one can argue that the functions y → Jy
(dz) have uniformly bounded variation in (Cc0,1 )∗ , modulo terms that vanish in still weaker norms, and this gives the necessary strong convergence. (The terms involving weaker norms force us to work with test functions that are C 2 instead of C 1 in much of the proof.) The remaining points follow quite directly from (ii) and the rectifiability criterion of Theorem 5.1, and from the two-dimensional results.

6 Dynamic Problems: Evolution of Vortex Filaments The stationary results, in particular the energy lower bounds and the compactness of the Jacobian are very useful tools in the asymptotic study of the evolution problems related to I
. In these note we only consider the parabolic equation, which is the gradient flow of I
u
t − ∆u
=

u
(1 − |u
|2 ), 2

t > 0, x ∈ IRn ,

(88)

x ∈ IRn ,

(89)

where the unknown function u
: IRn → IR2 satisfies an initial condition u
(0, x) = u
0 (x),

where u
0 is a given function. Here we do not consider the corresponding nonlinear Schr¨ odinger and the nonlinear heat equations. Instead we refer to the notes of Rubinstein [23] for an introduction and the description of formal results and to the paper of Colliander and Jerrard [7] for a rigorous study of the nonlinear, planar Schr¨ odinger equation. Also here we only consider the case n ≥ 3. Results for n = 2 are outlined in the Introduction. The asymptotic analysis of these equations initiated by the seminal paper of Rubinstein, Sternberg and Keller [24]. Much has been done for the scalar equation since then. We refer to the survey article of the author [30] and the references therein for the scalar equation, which is also called as the CahnAllen equation in the literature. The vector valued Ginzburg-Landau equation with u
: IRn → IRm is studied first by Struwe [32]. He considered this equation as a relaxation of the heat flow for harmonic maps. Under the assumption that original (not rescaled) energy is uniformly bounded in , he obtained deep partial regularity results. A monotonicity result is one of the interesting results of [32]. Our results differ from [32] in that we only consider the case u
∈ IR2 but we

226

Halil Mete Soner

allow for singularities to form and study the time evolution of the singular structures. Here we outline the results of [17] and [4]. This technique is also outlined in the lecture notes of the author [30] for the scalar equations. As discussed in the Introduction, in the following proof, we will use the compactness of the Jacobian to provide a shorter proof. 6.1 Energy identities Set e
:= e
(t, x) = E
(u
(t, x)). Recall that the energy density E
is given as E
(u) :=

1 (1 − |u|2 )2 |∇u|2 + . 2 42

By calculus and (88), (e
)t = −|u
t |2 + divp
,

(90)

∇e
= −p
+ div(∇u
⊗ ∇u
),

(91)

p
:= ∇u
u
t . To localize the energy estimates, let η ≥ 0 be a smooth compactly supported test function. Multiply (90) by η and (91) by ∇η and subtract the two identities. Then use the resulting identity to compute the time derivative of the integral of ηe
. The result is


d η e
= (ηt − ∆η) e
+ D2 η∇u
· ∇u
− η |u
t |2 . (92) dt Although we will not use it in these notes, let us mention that if we add the two identities instead of subtracting them, we obtain the following identity

d η e
= (ηt + ∆η) e
− D2 η∇u
· ∇u
(93) dt 
 
∇η · ∇u
2 |∇η · ∇u
|2   . − η ut − +  η η In [32], Struwe used the above identity with the special choice for η   1 |x − x0 |2 η(t, x) = , t < t0 , x ∈ IRn , exp − 4(t0 − t) (4(t0 − t))(n−m)/2 where t0 > 0 and x0 ∈ IRn are arbitrary, and m is the dimension of the range of u. The special case of (92) with η ≡ 1 yields the classical energy estimate,

t
e
(t, x) dx + |u
t |2 dxdt = e
(0, x) dx. (94) 0

Variational and Dynamic Problems for the Ginzburg-Landau Functional

227

6.2 Mean curvature flow and the distance function The distance and the square distance functions to a smooth manifold can be used to describe all the relevant geometric quantities. These functions were first used in [17] in the convergence proofs. Let {Γt }t∈[0,T ] be a smooth solution of co-dimension two mean curvature flow. Then the square distance function η(t, x) satisfies ηt − ∆η = −2,

on Γt ,

∇ [ηt − ∆η] = 0, D2 η ≤ In ,

on Γt ,

on whenever it is smooth.

(95) (96) (97)

The equation (96) is an observation of De Giorgi. It can be viewed as the definition of the codimension two mean curvature flow. We refer to [3] for information on the mean curvature flow in any codimension. However, (95), and (97) are the properties of any square distance function to a smooth codimension two manifold. Since Γt is smooth, η is smooth in a tubular neighborhood of Γt . (95) and (96) imply that in this neighborhood, |ηt − ∆η + 2| ≤ C η,

(98)

for some constant C. We extend η smoothly to all of [0, T ] × IRn so that the extension satisfies (97), (98), and η ≥ 0,

and η(t, x) = 0,

if and only if x ∈ Γt .

(99)

In (92) we choose η to be the square distance function, modified as above. Set
1
α (t) := η e
(t, x) dx, ln(1/) so that by (92), (97) and (98),

  d
1 1 α (t) ≤ C ηt e
− 2e
+ |∇u
|2 − |u
t |2 dx dt ln(1/) ln(1/)
1 ≤ C α
(t) − |u
t |2 dx. (100) ln(1/) 6.3 Convergence In this subsection, we prove the convergence of the solutions of 88 to smooth solutions of the mean curvature flow. This is first proved in [17], and here we follow a new approach using the compactness result on the Jacobians. The

228

Halil Mete Soner

convergence to weak solutions of the mean curvature flow is still not known. A first step in this direction is proved in [4]. Here we assume that the initial data concentrates on a smooth codimension two manifold Γ0 and that there exists a smooth solution {Γt }t∈[0,T ] of the codimension two mean curvature flow. then, we will prove that the energy measure concentrates on the smooth solution. Let
1
µt (V ) := E
(u
(t, x)) dx ln(1/) V be the rescaled Ginzburg-Landau energy. Precisely, we assume that u
0 is such that  µ
0 π Hn−2 Γ0 , in the weak∗ topology of Radon measure, where Hn−2 Γ0 is the n − 2 dimensional Hausdorff measure restricted to the codimension two manifold Γ0 , i.e, it is the surface area measure of Γ0 . A simple, consequence of this assumption is that C ∗ := sup µ
0 (IRn ) < ∞.


This together with the energy estimate (94) imply sup{ µ
t (IRn ) | t ∈ [0, T ],  > 0 } ≤ C ∗ . Using an argument due to Brakke (see [17] for details), this implies that there exists a subsequence, denoted by  again, such that µ
t



µt ,

t ∈ [0, T ],

where {µt }t∈[0,T ] stands for a family of nonnegative Radon measures.   Theorem 6.1. Suppose that Γ (t) t∈[t ,t ] is a collection of compact sets 0 1 which is a classical solution of the codimension two mean curvature flow. Let µt be a weak∗ limit of the rescaled energy measure µ
t . Then, µt ≥ πH1 Γt ,

t ∈ [0, T ],

and support µt = Γt ,

t ∈ [0, T ].

Proof. Since each connected component of the solution can be studied separately, without loss of generality, we may assume that Γt is conneceted with no boundary. 1. Let α
(t) be as in the previous subsection. Then by the convergence of µ
t ,

η µt (dx). lim α (t) = α(t) :=
→0

Variational and Dynamic Problems for the Ginzburg-Landau Functional

229

In view of (99) and our assumption on µ0 , α(0) = 0. Then by (100) and the Gronwall’s inequality α(t) ≤ α(0) eCt = 0. Since η is nonnegative, this implies that the support of µt is included in the zero set of η. Hence by (99), support µt ⊂ Γt ,

t ∈ [0, T ].

2. Fix t ∈ [0, T ] and let Jt
:= Ju
(t, ·). Since the rescaled Ginzburg-Landau energy µ
t (IRn ) is uniformly bounded in , we may apply the compactness result proved earlier. Then, on a subsequence denoted by  again, Jt
converges to a signed Radon measure Jt in the topology of (C 0,α )∗ for every α > 0. Moreover, d|Jt | ≤ 1, µt a.e., dµt where |Jt | is the total variation of the vector valued measure Jt and µt is the weak∗ limit of µ
t . Hence, by step 1, support Jt ⊂ support µt ⊂ Γt . 3. Since Jt
= dj(u
(t, ·)),

and

dJt
= 0

⇒

dJt = 0.

If we see Jt as a current, the above states that it has no boundary. By the previous step Jt is included in a smooth manifold Γt which has no boundary. In Lemma 6.1 below, we will show that these fact imply that the density of |Jt | on Γt d|Jt | Θt := dHn−2 Γt is constant. We postpone the proof of this result to Lemma 6.1 and complete the proof of this theorem. In view of our compactness result, this constant has to be an integer multiple of π, i.e., Θt ≡ nt π, Hn−2 Γt a.e., for some integer nt . 4. It suffices to show that nt ≡ 1 for t ∈ [0, T ]. Indeed this implies that |Jt | = π Hn−2 Γt . Since |Jt | ≤ µt ,

230

Halil Mete Soner

µt ≥ π Hn−2 Γt . and Γt ⊂ support µt . The opposite inclusion is proved in Step 1. 5. To prove that nt ≡ 1, consider the space-time Jacobian J
of u
on [0, T ] × IRn . The space-time Ginzburg-Landau energy is E
:=

 (1 − |u
|2 )2 1 |∇u
|2 + |u
t |2 + , 2 42

and in view of (94), 1 sup { ln(1/)





T


E
dx dt } < ∞. IRn

0

Hence we may apply our compactness result to J
, concluding that on a subsequence, denoted by  again, it converges to a Radon measure J ∗ . Moreover |J ∗ |([0, T ] × IRn ) ≤ µ([0, T ] × IRn ), where µ is the weak∗ limit of the space-time energy. Then,
T n µt (IRn ) dt + ν([0, T ] × IRn ), µ([0, T ] × IR ) = 0 ∗

where ν is the weak limit of ν
(dt × dx) :=

1 |u
|2 dt × dx. ln(1/) t

Let η and α be as in Step 1. By (92), and the properties of η,
d
1 α (t) = C α
(t) − η(t, x) |u
t |2 dx. dt 2 ln(1/) IRn By the Gronwall’s inequality α
(T ) ≤ α
(0) eCT − ≤ α
(0) eCT −

1 2 ln(1/) 1 2 ln(1/)




T

IRn

0







T

eC(T −t) η(t, x) |u
t |2 dx dt




η(t, x) ν
(dt × dx). 0

IRn

Since α
tends to zero as  approaches to zero. In the limit we obtain
T
η ν = 0. 0

IRn

Hence, the support of ν is included in the graph Γ of {Γt }t∈[0,T ]

Variational and Dynamic Problems for the Ginzburg-Landau Functional

231

Γ := { (t, x) ∈ [0, T ] × IRn | x ∈ Γt }. Therefore,

support J ∗ ⊂ Γ.

Using Lemma 6.1, we conclude that |J ∗ | = π Hn−1 Γ , and consequently |Jt | = π Hn−2 Γt ,

t ∈ [0, T ]. 2

In the following lemma we assume J is a Radon measure of the form in Theorem 5.2. In particular it satisfies (86)

φ(x)J(dx) = φ(x) · ν(x)Θ(x)Hm−2 (dx) ∀φ ∈ C 0 (IRm ; Λ2 IRm ). M

Lemma 6.1. Let J be as above. Further assume that M ⊂ IRn be a codimension two, smooth manifold with no boundary. Then, Θ is constant on M. Proof. Let J˜ be the same as J but with Θ ≡ 1, i.e.,

˜ φ(x)J(dx) = φ(x) · ν(x)Hm−2 (dx) ∀φ ∈ C 0 (IRm ; Λ2 IRm ). M

Since M has no boundary and since M is smooth, we directly calculate that dJ˜ = 0. We also know that dJ = 0. Using these two facts we calculate that ˜ = Θ d J˜ + d Θ ∧ J˜ 0 = d J = d [Θ J] ˜ = d Θ ∧ J. Hence ∇tan Θ = 0,

on M,

where ∇tan is the tangential derivative on M . Since M is connected, this implies that Θ is constant on M . 2

References [1] [2]

G. Alberti, S. Baldo and G. Orlandi, in preparation. L. Ambrosio and B. Kirchheim, Currents in metric spaces, Math. Ann., 318, 527-555 (2000).

232 [3] [4]

[5]

[6] [7]

[8] [9]

[10] [11] [12] [13] [14] [15] [16] [17]

[18] [19] [20] [21] [22] [23]

[24]

Halil Mete Soner L. Ambrosio and H.M. Soner, Level set approach to mean curvature flow in arbitrary codimension, J. Diff. Geometry, 43, 693-737 (1996). L. Ambrosio and H.M. Soner, A measure theoretic approach to higer codimension mean curvature flow, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 25, 27-48 (1997). F. Bethuel, A characterization of maps in H 1 (B 3 ; S 2 ) which can be approximated by smooth maps, Ann. Inst. H. Poincar´e Anal. Non Lin´eaire, 7, 269-286 (1990). F. Bethuel, H. Brezis and F. H´elein, Ginzburg Landau Vortices, Birkhauser, New-York, 1994. J.E. Colliander and R.L. Jerrard, Ginzburg-Landau vortices; weak stability and Schr¨ odinger equation dynamics, Journal d’Analyse Mathematique, 77, 129-205 (1999). G. Dal Maso, An Introduction to Γ -Convergence, Birkhauser, Boston, 1993. F. Demengel, Une caract´erisation des applications de W 1,p (B N , S 1 ) qui peuvent ˆetre approch´ees par des fonctions r´egulieres. C. R. Acad. Sci. Paris S´ er. I Math., 310, 553-557 (1990). M. Giaquinta, G. Modica and J. Soucek, Cartesian Currents in the Calculus of Variations I, Springer-Verlag, New York, 1998. M. Giaquinta, G. Modica and J. Soucek, Cartesian Currents in the Calculus of Variations II, Springer-Verlag, New York, 1998. R.L. Jerrard, Lower bounds for generalized Ginzburg-Landau functionals, SIAM Math. Anal., 30, 721-746 (1999). R.L. Jerrard and H.M. Soner, The Jacobian and the Ginzburg-Landau energy, Calc. Var. Partial Differential Equations, 14, 151-191 (2002). R.L. Jerrard and H.M. Soner, Functions of bounded higher variation, Indiana Univ. Math. J., 51, 645-677 (2002). R.L. Jerrard and H.M. Soner, Rectifiability of the distributional Jacobian for a class of functions, C. R. Acad. Sci. Paris S´ er. I Math., 329, 683-688 (1999). R.L. Jerrard and H.M. Soner, Dynamics of Ginzburg-Landau vortices, Arch. Rational Mech. Anal., 142, 185-206 (1998). R.L. Jerrard and H.M. Soner, Scaling limits and regularity for a class of Ginzburg-Landau systems, Ann. Inst. H. Poincar´e Anal. Non Lin´eairee, 16, 423-466 (1999). F.H. Lin, Some dynamical properties of Ginzburg-Landau vortices, Comm. Pure Appl. Math., 49, 323-359 (1996). F.-H. Lin, Solutions of Ginzburg-Landau equations and critical points of the renormalized energy, C.R. Acad. Sci. Paris, S´ erie I, 12, 599-622 (1995). L. Modica, The gradient theory of phase transitions and the minimal interface criterion, Arch. Rational Mech. Anal., 98, 123-142 (1987). L. Modica and S. Mortola, Il limite nella Γ -convergenza di una famiglia di funzionali elliptici, Boll. Un. Mat. Ital. A, 14, 526-529 (1977). L. Modica and S. Mortola, Un esempio di Γ convergenza, Boll. Un. Mat. Ital. B, 14, 285-299 (1977). J. Rubinstein, Six lectures on superconductivity. Boundaries, interfaces, and transitions (Banff, AB, 1995), 163-184, CRM Proc. Lecture Notes, 13, Amer. Math. Soc., Providence, RI, 1998. J. Rubinstein, P. Sternberg and J. Keller, Fast reaction, slow diffusion and curve shortening, SIAM J. Appl. Math., 49, 116-133 (1989).

Variational and Dynamic Problems for the Ginzburg-Landau Functional

233

[25] E. Sandier, Lower bounds for the energy of unit vector fields and applications, J. Funct. Anal., 152, 379-403 (1998). [26] E. Sandier, Ginzburg-Landau minimizers from IRn+1 into IRn and minimal connections, Indiana Univ. Math. J., 50, 1807-1844 (2001). [27] E. Sandier and S. Serfaty, Global minimizers for the Ginzburg-Landau functional below the first critical magnetic field, Ann. Inst. H. Poincar´e Anal. Non Lin´eaire, 17, 119-145 (2000). [28] L. Simon, Lectures on Geometric Measure Theory, Australian National University, 1984. [29] H.M. Soner, Ginzburg-Landau equation and motion by mean curvature. I. Convergence; II. Development of the initial interface, J. Geometric Analysis, 7, 437-475; 476-491 (1995). [30] H. M. Soner, Front propagation. Boundaries, interfaces, and transitions (Banff, AB, 1995), 185-206, CRM Proc. Lecture Notes, 13, Amer. Math. Soc., Providence, RI, 1998. [31] M. Struwe, On the asymptotic behavior of minimizers of the Ginzburg-Landau model in two dimensions, Differential Integral Equations, 7, 1613-1624 (1994). [32] M. Struwe, On the evolution of harmonic maps in higher dimensions, J. Diff. Geometry, 28, 485-502 (1988). [33] B. White, Rectifiability of flat chains, Annals of Mathematics, 150, 165-184 (1999).

List of Participants

1. Carlos Albuquerque, Univ. de Lisboa, Portugal [email protected] 2. Luigi Ambrosio, Scuola Normale Superiore, Pisa, Italy [email protected] 3. Rebekka Axthelm, Univ. Freiburg, Germany [email protected] 4. Michal Benes, Czech Tech. Univ. of Prague, Czech Rep. [email protected]fi.cvut.cz 5. Zsolt Bir´ o, CAR Inst. Hungarian Acad. Sci., Budapest, Hungary [email protected] 6. Elena Bonetti, Univ. Pavia, Italy [email protected] 7. Antonin Chambolle, Univ. Paris-Dauphine, France [email protected] 8. Pierluigi Colli, Univ. Pavia, Italy [email protected] 9. Armin Dahr, Univ. Bonn, Germany [email protected] 10. Klaus Deckelnick, Otto-von-Guericke-Universit¨ at Magdeburg, Germany [email protected] 11. Camillo De Lellis, Scuola Normale Superiore di Pisa, Italy [email protected] 12. Irina Denissova, IPME Russian Acad. Sci., St. Petersburg, Russia [email protected] 13. Lars Diening, Univ. Freiburg, Germany [email protected] 14. Fathi Dkhil, Univ. Cergy-Pontoise, France [email protected] 15. Gerhard Dziuk, Univ. Freiburg, Germany [email protected] 16. Angiolo Farina, Univ. Firenze, Italy [email protected]fi.it 17. Marcus Garvie, Univ. Durham, UK [email protected] 18. Marina Ghisi, Univ. Pisa, Italy [email protected]

236

19. Gianni Gilardi, Univ. Pavia, Italy [email protected] 20. Massimo Gobbino, Univ. Pisa, Italy [email protected] 21. C´esar Gon¸calves, Instituto Polit´ecnico da Guarda, Portugal [email protected] 22. Celine Grandmont, Univ. Paris-Dauphine, France [email protected] 23. Eurica Henriques, Univ. Tr´ as-os-Montes e Alto Douro, Portugal [email protected] 24. Viet Ha Hoang, Univ. Cambridge, UK [email protected] 25. Ryo Ikota, Univ. Tokyo, Japan [email protected] 26. Helge Kristian Jenssen, S.I.S.S.A., Trieste, Italy [email protected] 27. Moritz Kassmann, Univ. Bonn, Germany [email protected] 28. Omar Lakkis, Univ. Maryland, USA [email protected] 29. Stephen Langdon, Univ. of Durham, UK [email protected] 30. Irina Loginova, Royal Inst. Techn., Stockholm, Sweden [email protected] 31. Juergen Mehnert, Univ. Freiburg, Germany [email protected] 32. A. Meirmanov, Univ. da Beira Interior, Portugal [email protected] 33. Masayasu Mimura, Univ. Hiroshima, Japan [email protected] 34. Fernando Miranda, Univ. Minho, Portugal [email protected] 35. Maria Giovanna Mora, S.I.S.S.A., Trieste, Italy [email protected] 36. Massimiliano Morini, S.I.S.S.A., Trieste, Italy [email protected] 37. Gonoko Moussa, Univ. Nancy, France [email protected] 38. Piotr Mucha, Warsaw Univ., Poland [email protected] 39. Matteo Novaga, Univ. Pisa, Italy [email protected] 40. Hermenegildo Oliveira, Univ. Algarve, Portugal [email protected]

List of Participants

237

41. Emanuele Paolini, Scuola Normale Superiore di Pisa, Italy [email protected] 42. Peter Philip, Weierstrass Inst. App. Anal. Stoch., Berlin, Germany [email protected] 43. Mahmoud Qafsaoui, LAMFA-Univ. Picardie, France [email protected] 44. Vincenzo Recupero, Univ. Trento, Italy [email protected] 45. Elisabetta Rocca, Univ. Pavia, Italy [email protected] 46. Jos´e Francisco Rodrigues, Univ. de Lisboa, Portugal [email protected] 47. Mathias R¨ oger, Univ. Bonn, Germany [email protected] 48. Raluca Rusu, Univ. Freiburg, Germany [email protected] 49. Giulio Schimperna, Univ. Pavia, Italy [email protected] 50. Daniel Sevcovic, Comenius Uiversity, Bratislava [email protected] ´ 51. Israel Michael Sigal, Institute des Hautes Etudes Scientifiques, France [email protected] 52. Boyan Sirakov, Univ. Paris VI, France [email protected] 53. Vsvolod A. Solonnikov, Steklov Math. Inst., St. Petersburg, Russia [email protected] 54. Halil Mete Soner, Ko¸c University, Istanbul, Turkey [email protected] 55. Ulisse Stefanelli, I.M.A.T.I.-C.N.R., Pavia, Italy [email protected] 56. Nicolas van Goethem, Univ. de Pisa, Italy [email protected] 57. Remi Weidenfeld, Univ. Paris-Sud, France [email protected] 58. Sandra Wieland, Univ. Bonn, Germany [email protected]

LIST OF C.I.M.E. SEMINARS

1954

1. Analisi funzionale 2. Quadratura delle superficie e questioni connesse 3. Equazioni differenziali non lineari

1955

4. 5. 6. 7.

1956

8. 9. 10. 11.

Teorema di Riemann-Roch e questioni connesse Teoria dei numeri Topologia Teorie non linearizzate in elasticit` a, idrodinamica, aerodinamic Geometria proiettivo-differenziale Equazioni alle derivate parziali a caratteristiche reali Propagazione delle onde elettromagnetiche Teoria della funzioni di pi` u variabili complesse e delle funzioni automorfe Geometria aritmetica e algebrica (2 vol.) Integrali singolari e questioni connesse Teoria della turbolenza (2 vol.)

C.I.M.E " " " " " " " " " " " " "

1957

12. 13. 14.

1958

15. Vedute e problemi attuali in relativit` a generale 16. Problemi di geometria differenziale in grande 17. Il principio di minimo e le sue applicazioni alle equazioni funzionali 18. Induzione e statistica 19. Teoria algebrica dei meccanismi automatici (2 vol.) 20. Gruppi, anelli di Lie e teoria della coomologia

" " "

1960

21. Sistemi dinamici e teoremi ergodici 22. Forme differenziali e loro integrali

" "

1961

23. Geometria del calcolo delle variazioni (2 vol.) 24. Teoria delle distribuzioni 25. Onde superficiali

" " "

1962

26. Topologia differenziale 27. Autovalori e autosoluzioni 28. Magnetofluidodinamica

" " "

1963

29. Equazioni differenziali astratte 30. Funzioni e variet` a complesse 31. Propriet` a di media e teoremi di confronto in Fisica Matematica

" " "

1959

" " "

240 1964

1965

1966

1967

1968

LIST OF C.I.M.E. SEMINARS 32. 33. 34. 35. 36. 37. 38.

Relativit` a generale Dinamica dei gas rarefatti Alcune questioni di analisi numerica Equazioni differenziali non lineari Non-linear continuum theories Some aspects of ring theory Mathematical optimization in economics

39. 40. 41. 42. 43. 44.

Calculus of variations Economia matematica Classi caratteristiche e questioni connesse Some aspects of diffusion theory Modern questions of celestial mechanics Numerical analysis of partial differential equations Geometry of homogeneous bounded domains Controllability and observability Pseudo-differential operators Aspects of mathematical logic

45. 46. 47. 48.

C.I.M.E " " " " " " Ed. Cremonese, Firenze " " " " " " " " "

49. Potential theory 50. Non-linear continuum theories in mechanics and physics and their applications 51. Questions of algebraic varieties 52. Relativistic fluid dynamics 53. Theory of group representations and Fourier analysis 54. Functional equations and inequalities 55. Problems in non-linear analysis 56. Stereodynamics 57. Constructive aspects of functional analysis (2 vol.) 58. Categories and commutative algebra

" "

1972

59. Non-linear mechanics 60. Finite geometric structures and their applications 61. Geometric measure theory and minimal surfaces

" " "

1973

62. Complex analysis 63. New variational techniques in mathematical physics 64. Spectral analysis 65. Stability problems 66. Singularities of analytic spaces 67. Eigenvalues of non linear problems

" "

1975

68. Theoretical computer sciences 69. Model theory and applications 70. Differential operators and manifolds

" " "

1976

71. Statistical Mechanics 72. Hyperbolicity 73. Differential topology

1977

74. Materials with memory 75. Pseudodifferential operators with applications 76. Algebraic surfaces

1969

1970

1971

1974

" " " " " " " "

" " " "

Ed. Liguori, Napoli " " " " "

LIST OF C.I.M.E. SEMINARS 1978

77. Stochastic differential equations 78. Dynamical systems

1979

79. Recursion theory and computational complexity 80. Mathematics of biology

" "

1980

81. Wave propagation 82. Harmonic analysis and group representations 83. Matroid theory and its applications

" " "

1981

84. Kinetic Theories and the Boltzmann Equation 85. Algebraic Threefolds 86. Nonlinear Filtering and Stochastic Control

(LNM 1048) (LNM 947) (LNM 972)

1982

87. Invariant Theory 88. Thermodynamics and Constitutive Equations 89. Fluid Dynamics

(LNM 996) (LN Physics 228) (LNM 1047)

" " "

1983

90. Complete Intersections 91. Bifurcation Theory and Applications 92. Numerical Methods in Fluid Dynamics

(LNM 1092) (LNM 1057) (LNM 1127)

" " "

1984

93. Harmonic Mappings and Minimal Immersions 94. Schr¨ odinger Operators 95. Buildings and the Geometry of Diagrams

(LNM 1161) (LNM 1159) (LNM 1181)

" " "

1985

96. Probability and Analysis 97. Some Problems in Nonlinear Diffusion 98. Theory of Moduli

(LNM 1206) (LNM 1224) (LNM 1337)

" " "

Ed. Liguori, Napoli & Birkh¨ user "

Springer-Verlag " "

1986

99. Inverse Problems 100. Mathematical Economics 101. Combinatorial Optimization

(LNM 1225) (LNM 1330) (LNM 1403)

" " "

1987

102. Relativistic Fluid Dynamics 103. Topics in Calculus of Variations

(LNM 1385) (LNM 1365)

" "

1988

104. Logic and Computer Science 105. Global Geometry and Mathematical Physics

(LNM 1429) (LNM 1451)

" "

1989

106. Methods of nonconvex analysis 107. Microlocal Analysis and Applications

(LNM 1446) (LNM 1495)

" "

1990

108. Geometric Topology: Recent Developments 109. H∞ Control Theory 110. Mathematical Modelling of Industrial Processes

(LNM 1504) (LNM 1496) (LNM 1521)

" " "

1991

111. Topological Methods for Ordinary Differential Equations 112. Arithmetic Algebraic Geometry 113. Transition to Chaos in Classical and Quantum Mechanics 114. Dirichlet Forms 115. D-Modules, Representation Theory,and Quantum Groups 116. Nonequilibrium Problems in Many-Particle Systems

(LNM 1537)

"

(LNM 1553) (LNM 1589)

" "

(LNM 1563) (LNM 1565)

" "

(LNM 1551)

"

1992

241

242

LIST OF C.I.M.E. SEMINARS

1993

117. Integrable Systems and Quantum Groups 118. Algebraic Cycles and Hodge Theory 119. Phase Transitions and Hysteresis

(LNM 1620) (LNM 1594) (LNM 1584)

Springer-Verlag " "

1994

120. Recent Mathematical Methods in Nonlinear Wave Propagation 121. Dynamical Systems 122. Transcendental Methods in Algebraic Geometry 123. Probabilistic Models for Nonlinear PDE’s 124. Viscosity Solutions and Applications 125. Vector Bundles on Curves. New Directions

(LNM 1640)

"

(LNM (LNM (LNM (LNM (LNM

1609) 1646) 1627) 1660) 1649)

" " " " "

(LNM 1684)

"

(LNM 1713)

"

(LNM 1656) (LNM 1714) (LNM 1697)

" " "

(LNM 1716) (LNM 1776) (LNM 1740) (LNM 1784) to appear

" " " " "

(LNM 1715)

"

(LNM 1734)

"

(LNM 1739)

"

(LNM 1804)

"

to appear

"

(LNM 1775) to appear

" "

to appear (LNM 1819) (LNM 1812) to appear to appear to appear (LNM 1802) (LNM 1813) to appear

" " " " " " " " "

to appear to appear to appear

" " "

1995

1996

1997

1998

1999

2000

2001

2002

126. Integral Geometry, Radon Transforms and Complex Analysis 127. Calculus of Variations and Geometric Evolution Problems 128. Financial Mathematics 129. Mathematics Inspired by Biology 130. Advanced Numerical Approximation of Nonlinear Hyperbolic Equations 131. Arithmetic Theory of Elliptic Curves 132. Quantum Cohomology 133. Optimal Shape Design 134. Dynamical Systems and Small Divisors 135. Mathematical Problems in Semiconductor Physics 136. Stochastic PDE’s and Kolmogorov Equations in Infinite Dimension 137. Filtration in Porous Media and Industrial Applications 138. Computational Mathematics driven by Industrial Applications 139. Iwahori-Hecke Algebras and Representation Theory 140. Theory and Applications of Hamiltonian Dynamics 141. Global Theory of Minimal Surfaces in Flat Spaces 142. Direct and Inverse Methods in Solving Nonlinear Evolution Equations 143. Dynamical Systems 144. Diophantine Approximation 145. Mathematical Aspects of Evolving Interfaces 146. Mathematical Methods for Protein Structure 147. Noncommutative Geometry 148. Topological Fluid Mechanics 149. Spatial Stochastic Processes 150. Optimal Transportation and Applications 151. Multiscale Problems and Methods in Numerical Simulations 152. Real methods in Complex and CR Geometry 153. Analytic Number Theory 154. Imaging

LIST OF C.I.M.E. SEMINARS 2003

155. 156. 157. 158.

Stochastik Methods in Finance Hyperbolik Systems of Balance Laws Sympletic 4-Manifolds and Algebraic Surfaces Mathematical Foundation of Turbulent Viscous Flows

announced announced announced announced

243

Fondazione C.I.M.E. Centro Internazionale Matematico Estivo International Mathematical Summer Center http://www.math.unifi.it/∼cime [email protected]fi.it

2003 COURSES LIST Stochastic Methods in Finance July 6–13, Cusanus Akademie, Bressanone (Bolzano) Joint course with European Mathematical Society Course Directors: Prof. Marco Frittelli (Univ. di Firenze), [email protected]fi.it Prof. Wolfgang Runggaldier (Univ. di Padova), [email protected]

Hyperbolic Systems of Balance Laws July 14–21, Cetraro (Cosenza) Course Director: Prof. Pierangelo Marcati (Univ. de L’Aquila), [email protected]

Symplectic 4-Manifolds and Algebraic Surfaces September 2–10, Cetraro (Cosenza) Course Directors: Prof. Fabrizio Catanese (Bayreuth University) Prof. Gang Tian (M.I.T. Boston)

Mathematical Foundation of Turbulent Viscous Flows September 1–6, Martina Franca (Taranto) Course Directors: Prof. M. Cannone (Univ. de Marne-la-Vall´ee) Prof.T. Miyakawa (Kobe University)