A Dissertation entitled
Sobolev Gradient Semi-flows & Applications to Nonlinear Problems
by Ramesh Karki
Submitted to the Graduate Faculty as partial fulfillment of the requirements for the Doctor of Philosophy Degree in Mathematics
Dr. Alessandro Arsie, Committee Chair
Dr. Denis A. White, Committee Member
Dr. Franco Cardin, Committee Member
Dr. Gregory S. Spradlin, Committee Member
Dr. Mao-Pei Tsui, Committee Member
Dr. Patricia R. Komuniecki, Dean College of Graduate Studies
The University of Toledo May 2015 Copyright 2015, Ramesh Karki
This document is copyrighted material. Under copyright law, no parts of this document may be reproduced without the expressed permission of the author. An Abstract of Sobolev Gradient Semi-flows & Applications to Nonlinear Problems by Ramesh Karki
Submitted to the Graduate Faculty as partial fulfillment of the requirements for the Doctor of Philosophy Degree in Mathematics The University of Toledo May 2015
We are interested in solving nonlinear pseudo-differential equations (in particular, partial differential equations as well) involving fractional powers of uniformly ellip- tic self-adjoint operators of order two with suitable smoothness conditions on the coefficients subject to given (Dirichlet or periodic) boundary conditions. Under the stronger assumptions, we are interested in studying solutions in a special class whose elements satisfy non-selfintersecting property and have bounded distance from a given hyperplane, since such solutions are the analogue for Aubrey-Mather sets for ODEs and leaves of minimal foliations or laminations for PDEs . To solve such a ΨDE, we will start by introducing an energy type functional whose Euler-Lagrange equation is the pseudo-differential equation itself. As we seek to minimize this functional, we will introduce the Sobolev gradient of the functional as an element of a suitable Sobolev space and then we consider the gradient descent equation subject to appropriate ini- tial and boundary conditions. The equilibrium solutions of this Sobolev gradient descent equation are the critical points we are looking for.
Now the first step of our work will be to construct a semi-flow corresponding to the aforementioned initial-boundary value problem. So we will prove the existence, uniqueness, regularity, and comparison properties related to the semi-flow. Then
iii the next step will be to analyze the convergence of this semi-flow to an equilibrium solution to this initial-boundary value problem. In our work, we will adapt two methods: analytical method and numerical method. We apply various analytical tools to establish the general results and numerical tools to study concrete solutions of particular pseudo-differential or partial differential equations.
iv Dedicated to
“My Professors, Teachers, Family, and Friends” who have directly or indirectly supported me in my academic life Acknowledgments
I would like to sincerely thank my Ph. D. adviser Prof. Alessandro Arsie for his continuous advice and tireless support to develop foundation in my research area and successfully establish main results and many other technical results included in this dissertation. I would like to thank my dissertation committee members Dr. Denis
A. White, Dr. Mao-Pei Tsui, Dr. Franco Cardin and Dr. Gregory Spradlin for their careful reading of this manuscript, feedback and suggestions.
I would also like to thank Dr. Timothy Blass and Prof. Rafael de lLave for their support in many ways. Their kind responses of my questions in emails and in person during several meetings, their help on providing notes, papers and books, and giving comments were instrumental to proceed my research work ahead.
I am particularly indebted to my good friends and professors Dr. Akaki Tikaradze,
Dr. Vani Cheruvu, Dr. Santosh Kandel, Dr. Harihar Khanal, who have been help- ing me discuss on several topics and problems, giving me valuable suggestions and encouragements to obtain my academic achievements by several means.
Finally, I would like to express my sincere thanks to my family, many other friends, colleagues, and professors who have influenced directly and indirectly in my academic life so that their support has immensely helped me obtain my academic achievements.
Ramesh Karki
vi Contents
Abstract iii
Acknowledgments vi
Contents vii
List of Abbreviations x
List of Symbols xi
Preface xii
1 Introduction 1
2 Some Preliminaries 5
2.1 Euler-Lagrange Equation ...... 5
2.2 Sobolev Spaces ...... 7
2.2.1 When Ω = Td ...... 7 2.2.2 When Ω is an open bounded domain ...... 10
2.3 Sobolev Gradients ...... 11
2.4 Sobolev Gradient Descent Equations ...... 14
2.5 Semigroups And Fractional Powers of Linear Operators ...... 16
2.6 The Spectral Property For An Unbounded Self-adjoint Operator On A
Hilbert Space ...... 22
2.7 Main Results ...... 23
vii 2.7.1 Existence, Uniqueness & Regularity ...... 23
2.7.2 The Comparison Principles ...... 24
2.7.3 Applications To Semilinear ΨDEs ...... 25
2.7.4 Numerical Solutions To Some ΨDEs ...... 25
3 Characterization of Sobolev Spaces and Fractional Powers of Oper-
ators 27
3.1 Characterization of L2(Ω)...... 27
3.2 Characterizations of Hα(Ω), Hαβ(Ω) and Fractional Powers of Oper-
ators on L2(Ω)...... 28
3.3 Fractional Powers of Operators on Hαβ(Ω)...... 30
4 Constructions of Semigroups, Comparison, Regularity & Bounded-
ness 32
4.1 Constructions of Semigroups and Comparison ...... 33
4.2 Regularity and Boundedness of e−tAα ...... 37
4.3 Constructions of Semigroups and Comparison, revisited ...... 41
4.4 Regularity and Boundeness Properties of etL ...... 46
5 The Proofs of Main Results 50
5.1 The Proofs of Main Theorems Under L2-gradient ...... 50
5.1.1 Some Preliminaries ...... 50
5.1.2 The Proof of Theorem 2.7.1 ...... 54
5.1.3 The Proof of Theorem 2.7.3 ...... 58
5.2 The Proofs of Main Theorems Under Hαβ-gradient ...... 62
5.2.1 Some Preliminaries (Revisited) ...... 62
5.2.2 The Proof of Theorem 2.7.2 ...... 65
5.2.3 The Proof of Theorem 2.7.4 ...... 70
viii 6 Applications to Semilinear Pseudo Differential Equations 75
6.1 Nonself-intersecting Solutions ...... 76
6.2 A Generalization of Minimal Foliations (or Minimal Laminations) &
Invariant Tori (or Aubry-Mather Sets) ...... 77
6.3 The Proof of Theorem 2.7.5 ...... 82
6.4 Regularity in Some Special Cases ...... 91
6.5 Numerical Solutions to Some Semilinear PDEs and ΨDEs ...... 92
References 100
A Pseudo-differential Operators 103
B MATLAB Programmings 107
B.1 Code for Example 6.5.1 ...... 107
ix List of Abbreviations
E-L ...... Euler-Lagrange ACP ...... abstract Cauchy aroblem IVP ...... initial value problem IBVP ...... initial-boundary value problem ΨDE ...... pseudo-differential equation PDE ...... partial differential equation ODE ...... ordinary differetial equation FFT ...... fast Fourier transform DFT ...... discrete Fourier transform
x List of Symbols
R ...... The set of real numbers N ...... The set of positive integers NTd ...... The torus or N copies of [0,N]d or Rd/NZd
C∞(Ω)..... The space of smooth functions in Ω ∞ Cc (Ω)..... The space of smooth functions with compact supports in Ω S(R)...... The Schwarz space of rapidly decreasing smooth functions 0 ∞ D (Ω)...... The space of distributions or dual of Cc (Ω) S0(R)...... The space of tempered distributions or dual of S(R) L∞(Ω)...... The space of functions whose essentialsup norm is finite L2(Ω)...... The space of functions (more generally distributions) which are square integrable over Ω and satisfy either periodic or Dirichlet boundary conditions Hk(Ω)...... Sobolev space of order k consisting of distributions in L2(Ω) whose weak partial derivatives of orders ≤ k, k ∈ Z+ are in L2(Ω) Hs(Ω)...... Fractional order Sobolev space of order s Hαs(Ω)..... Fractional order Sobolev space of order αs
Inner Products( ., .)L2 = (., .)L2(Ω) , (., .)αβ = (., .)Hαβ (Ω) , (., .)αs = (., .)Hαs(Ω) , (., .)s = (., .)Hs(Ω) , (., .)k = (., .)Hk(Ω) , (., .)0 = (., .)H0(Ω)
Norms ...... k.k∞ = k.kL∞(Ω), k.kL2 = k.kL2(Ω), k.kαβ = k.kHαβ (Ω), k.kαs = k.kHαs(Ω), k.ks = k.kHs(Ω), k.kk = k.kHk(Ω), k.k0 = k.kH0(Ω)
L(H)...... The Banach space of all bounded linear operators on Hilbert space H under supnorm of operators.
α ∂|α| d D ...... = for multi-index α = (α1, . . . , αd) ∈ N with |α| = α1 + ∂xα1 ...∂xαd ∂αi th ··· + αd, where each is αi weak partial derivative. ∂xαi
α α1 αd d |x| ...... = x1 ...... xn for multi-index α = (α1, . . . , αd) ∈ N with |α| = α1 + ··· + αd.
xi Preface
The goal of this dissertation work is to give a brief presentation of Sobolev gradi- ents and to emphasize their applications on the theory of partial differential equations and pseudo-differential equations. We explore in depth the solutions to such equa- tions and some interesting behavior of the solutions using powerful tools from some of the rich theories of analysis such as the abstract semigroup theory, the spectral theory of linear unbounded self-adjoint operator on a Hilbert space, the theory of gen- eralized functions, functional analysis, and theory of partial differential equations and pseudo-differential equations. It, further, highlights on some numerical applications of these equations and opens up some problems in the future directions.
xii Chapter 1
Introduction
In this chapter we are going to discuss briefly what our main goal is, how we set
up problem, and what we do to solve the problem we set up. We also discuss about
what we prove so far, work in progress and the possible extensions of our problem or
some future projects.
Here our main interest is to solve certain types of nonlinear partial differential
equations (PDEs) and pseudo-differential equations (ΨDEs),
look for some special classes of solutions and investigate various properties. We
will investigate these equations and their solutions via analytical method as well as
numerical method.
To deal with each of these nonlinear PDEs and ΨDEs we consider in section 2.1 a
certain energy type functional Eα defined on an infinite dimensional Hilbert space H, namely a Sobolev space Hs(Ω), s ≥ 0, so that the equation satisfied by its critical points is the given PDE or ΨDE. We discuss these Sobolev spaces in section 2.2 and characterize them in sections 3.1 and 3.2. As the critical points of Eα are of our interest, we need to construct a suitable gradient of the functional as an element
of H. Such a gradient, denoted by ∇HEα(u), is a counterpart of the usual gradient
Pd ∂g 1 d ∇g = i=1 of a real valued C function g defined on the Euclidean space R . This ∂xi gradient needs to have the property that the solutions to the corresponding gradient
1 equation ∇HEα(u) = 0 are the critical points of the functional Eα. We consider H as a Sobolev space, introduce such a gradient as an element of the Sobolev space in section 2.3 and also give the formula for it there. Our method of solving a PDE or
ΨDE by finding critical points of the functional Eα is quite different from the method of solving a PDE by finding minimizers of an energy functional using the so called direct method of calculus of variations.
By the direct method of calculus of variations, a variational problem is solved by
finding minimizers of an energy functional. These minimizers with enough smoothness are the solutions to a PDE, which we call Euler-Lagrange equation for the energy functional. But unlike this method of finding the minimizers of Eα directly, we look for functions u which solve the gradient equation ∇HEα(u) = 0. We will show that this equation is in fact equivalent to the Euler-Lagrange equation of Eα. In section 2.2 we will introduce fractional order Sobolev spaces and continue them in Chapter 2 where we characterize them with the help of spectral theory of unbounded self-adjoint operator on a Hilbert space. In section 2.3 we define Sobolev gradients of Eα and give formulas for them.
We consider in section 2.4 the gradient descent equation
∂u = −∇ E (u) ∂t H α subject to the initial condition u(0,.) = u0(.) in appropriate spaces and the periodic or Dirichlet boundary conditions so that this initial-boundary value problem is well- posed. As we are seeking for finding functions u that solve the gradient equation
∇HEα(u) = 0, we need to look for equilibrium solutions to this initial-boundary value problem. To solve this IBVP, we recast it into the corresponding abstract
Cauchy problem (ACP) du = −∇HEα(u) dt (1.0.1) u[0] = u0
2 on the Sobolev space whose elements are accompanied by either of the above boundary conditions. The main part of this thesis contributes for proving the existence and uniqueness of solution to this ACP, improving regularity of the solution, establishing the comparison principle related to the solution, and applying these results to study the convergence of the solution under special circumstances.
In section 2.5 and section 2.6 we briefly discuss about the semigroup theory, some important results from spectral theory of unbounded self-adjoint operators on a Hilbert space and discuss the fractional powers of the self-adjoint operators (also in section 3.3). In section 2.7 we state some of the main original results we proved in this thesis such as the existence, uniqueness, regularity and comparison principle for the solution to the above ACP and the existence of solutions to the Euler-Lagrange equation of Eα in a class Bω. We will briefly introduce this class in this chapter and discuss about it later in Chapter 6.
Since the proof of the main results rely heavily on semigroup theory, throughout
Chapter 4 we construct various semigroups generated by a self-adjoint uniformly elliptic operator and its fractional powers on Sobolev spaces and discuss comparison, regularity and boundedness properties of these semigroups.
In Chapter 5 we first define a mild solution to ACP (1.0.1) as an integral form of the ACP using the semigroups constructed in Chapter 4. We prove that this solution is a globally attracting fixed point of a contraction map by using Banach
Fixed Point Theorem. Then we improve its regularity and complete the proof of existence, uniqueness and regularity results. Next we adapt the iteration method to prove the comparison principle by proving that these iterations satisfy the comparison principle and converge to the corresponding fixed points.
In Chapter 6 as another original main result of this dissertation, we prove the existence of solutions to the Euler Lagrange equation of Eα in the class Bω when ω is
d d a vector in Q . The class Bω consists of functions u : R → R with the properties that
3 the graph of each such function when projected on a torus does not intersect itself
(nonself-intersecting property) and that the hypersurface represented by xd+1 = u(x) has bounded distance from the hyperplane xd+1 = ω.x. We use the main results- the existence and uniqueness theorem, the regularity theorem and the comparison prin- ciple mentioned above to prove that the solution u(t, x) to the IBVP corresponding to ACP (1.0.1) starting at a point in Bω always remains in Bω. Moreover, for such a solution u(t, x) there is an increasing sequence of times {tn} approaching ∞ such that u(tn, x) converges to uω(x) as tn approaches ∞ and uω satisfies the Euler-Lagrange equation. But since Bω is closed under this convergence, uω is in Bω. In this way we prove the existence of solutions to the Euler-Lagrange equation of Eα in the class Bω.
The work of investigating whether the analogous result holds when ω is in Rd − Qd is in progress.
Our interest for studying the solutions to semilinear pseudo-differential equations
(Euler-Lagrange equations for Eα) in the class Bω stems from the study of minimal fo- liations or laminations for variational problems for PDEs and study of Aubrey-Mather sets for ODEs. Such a minimal foliation for a variational problem for PDE is a con- tinuous one parameter family of smooth hypersurfaces as graphs of functions, called leaves of the foliation, covering the fundamental domain and the function representing each leaf satisfies nonself-intersecting property and is a solution to the correspond- ing Euler-Lagrange equation. Aubrey-Mather sets arise when studying the stability problem for dynamical systems, especially for the systems of two degree of freedom for which such a problem is about constructing two dimensional invariant tori (a part of KAM theory). The solutions to ΨDE in the class Bω have similar properties as compared with the leaves of minimal foliations. We will briefly discuss these minimal foliations or laminations (foliations with gaps) and Aubrey-Mather sets in Chapter 6.
To know more about the minimal foliations or laminations, one can read [3, 11, 12, 13], and to know more about Aubrey-Mather sets, one can read [14].
4 Chapter 2
Some Preliminaries
In the course of proving our original results we apply various tools from abstract semigroup theory on a Hilbert space, spectral theory of linear unbounded self-adjoint operator on a Hilbert space, and many other concepts from functional analysis, dis- tribution theory, and theory of partial differential equations and pseudo-differential equations.
Now we are going to discuss briefly some topics and key concepts which we fre- quently use in studying our original results in this entire dissertation.
2.1 Euler-Lagrange Equation
Let Ω be an open bounded domain in Euclidean space Rd with smooth boundary
∂Ω or a closed manifold such as a torus NTd. Let f(x, y) be a bounded continuous real-valued function on Ω¯ × R such that its continuous first partial derivative with ¯ respect to y, fy, exists in R for all x ∈ Ω and is bounded on Ω × R. We are interested in finding solutions of the following semi-linear ΨDE
Aαu + f(x, u) = 0, x ∈ Ω (2.1.1)
subject to Dirichlet boundary conditions u|∂Ω = 0 or periodic boundary conditions
5 u(x + ej) = u(x) ∀ ej = (0,..., 1,..., 0), j = 1, . . . , d, where α ∈ (0, 1], and A is a linear, self-adjoint and uniformly elliptic operator of order 2 on L2(Ω), the space of square integrable functions satisfying either of the boundary conditions, given by
d X (Au)(x) = − ((aij(x)uxi (x))xj , x ∈ Ω (2.1.2) i,j=1
∞ with the properties that aij ≡ aji ∀ i, j = 1, 2 . . . , d and aij ∈ C (Ω), and
d 2 X 2 d θ1|ξ| ≤ aij(x)ξiξj ≤ θ2|ξ| ∀ x ∈ Ω, ∀ ξ ∈ R . (2.1.3) i,j=1
for some positive real numbers θ1, θ2.
The semi-linear ΨDE (2.1) is the Euler-Lagrange equation of the functional Eα,
Z 1 α/2 2 Eα(u) = |A u| + F (x, u) dx, α ∈ (0, 1] (2.1.4) Ω 2
R y ∞ defined on a Sobolev space, where F (x, y) = 0 f(x, z) dz. To see this, if φ ∈ Cc (Ω) then
d E (u + τφ)| dτ α τ=0 Z d n1 α/2 α/2 o = (A (u + τφ),A (u + τφ))L2(Ω) + F (x, u + τφ) dx |τ=0 dτ 2 Ω Z α/2 α/2 ∂ α/2 =(A u, A φ)L2(Ω) + F (x, u + τφ)|τ=0 dx (∵ A is linear) Ω ∂τ Z α α/2 =(A u, φ)L2(Ω) + f(x, u)φ dx (∵ A is self- adjoint) Ω α =(A u + f(x, u), φ)L2(Ω) = 0,
α ∞ 2 implies that A u + f(x, u) = 0 by the denseness of Cc (Ω) in L (Ω).
We will see in section 2.4 that (2.1) is equivalent to each of ∇αβEα(u) = 0 and
∇L2 Eα(u) = 0 for each critical point u of Eα, where ∇αβEα(u) and ∇L2 Eα(u) are
6 αβ 2 H (Ω)-gradient and L (Ω)-gradient of Eα at u respectively. This is the reason why we focus on studying the solutions to these gradient equations in order to solve (2.1).
2.2 Sobolev Spaces
Following [2, 9, 12, 17, 22, 24], first we recall definitions of some fractional order
Sobolev spaces and then introduce new Sobolev spaces.
We start with recalling the following Sobolev spaces as a Hilbert space with the given inner product:
k 2 α 2 H (Ω) = {u ∈ L (Ω): D u ∈ L (Ω) ∀ |α| ≤ k}, k ∈ N,
where Ω is open bounded domain or a torus or all of Rd and Dα is a partial derivative of order α in the distributional sense, with inner product and norm
1/2 X α ν n X α 2 o (u, v)k = (D u, D v)L2 , kukk = kD ukL2 and |α|≤k |α|≤k
s d 0 d 2 s/2 2 d H (R ) = {u ∈ S (R ) : (1 + |ξ| ) uˆ ∈ L (R )}, s ∈ R,
with inner product and norm
2 s/2 2 s/2 2 s/2 (u, v)Hs( d) = ((1 + |ξ| ) u,ˆ (1 + |ξ| ) vˆ)L2( d), kuk s d = (1 + |ξ| ) uˆ 2 d . R R H (R ) L (R )
2.2.1 When Ω = Td
In this case, we always consider L2(Ω) together with the usual L2-inner product
as the space of all square integrable functions that satisfy the periodic boundary
conditions.
7 We discuss various fractional order Sobolev spaces which are Hilbert spaces with corresponding inner products (see [1]).
s 0 X 2 s/2 ij.x 2 s H (Ω) = {u ∈ D (Ω): (1 + |j| ) uˆ(j)e ∈ L (Ω)} = D(Λ ), s ∈ R j∈2πZd where Λsu(x) := P (1 + |j|2)s/2uˆ(j)eij.x for u ∈ D0(Ω), with inner product and j∈2πZd norm
s s X 2 s X 2 s 2 1/2 (u, v)s = (Λ u, Λ v)L2 = (1+|j| ) uˆ(j)vˆ(j), kuks = { (1+|j| ) |uˆ(j)| } j∈2πZd j∈2πZd