Scattering Theory
Sean Harris
October 2016
A thesis submitted for the degree of Bachelor of Philosophy (Honours) of the Australian National University
Declaration
The work in this thesis is my own except where otherwise stated. The first four chapters are largely expository, covering background material. The final two chapters contain mostly new theorems (to the best of my knowledge) proven by myself, with some help by others such as Professor Andrew Hassell.
Sean Harris
Acknowledgements
Most of all I would like to thank my supervisor Professor Andrew Hassell, for developing my interest in this topic, always finding time to meet with me to discuss issues, and being generally great. I am also incredibly grateful to Dr Melissa Tacy for helping me draft this thesis while Andrew was absent. Thanks also goes to the many lecturers I have met leading up to this moment, for nourishing my interest in mathematics. My fellow Dungeon-dwelling honours students have also been incredibly helpful throughout the year. Special mention to Hugh McCarthy, for always being available to chat about any problems I was having (both mathematical and otherwise). A few more people to mention: my tutorial students, for giving me some reprieve from study; the staff at the ANU Food Co-op, for the great lunches (I wish I had discovered them earlier); and the many green envelopes and the single red envelope, for keeping me going. I guess I should thank my parents too.
v
Abstract
Scattering theory studies the comparison between evolution obeying “free dynamics” and evolution obeying some “perturbed dynamics”. The asymptotic nature of free and perturbed evolution are compared to determine properties of the perturbation. A brief introduction to scattering theory and inverse scattering problems is given in Chapter3, after covering some relevant analysis concepts and the construction of Laplacian and Dirichlet to Neumann operators in Chapters1 and2. The spectral duality result of [EP95] is as follows.
Theorem ([EP95], Main Theorem). The following are equivalent:
2 1. −∆D has an M-fold degenerate eigenvalue k0
2. As k ↑ k0, exactly M eigenphases θj(k) of SD(k) converge to π from below.
In Chapter4 this result is explained both mathematically and intuitively, and the significance of the form of the second equivalent statement is examined. Possible generalisations of the main theorem of [EP95] to potential scattering are then inves- tigated in Chapter5. The Cayley transform generalisation, as explained in Section 5.3, suggests studying an object which is given the name transmission operator. The transmission operator displays qualities similar to the scattering matrix and so is investigated further in Chapter6. Many results about the spectrum of the transmission operator are proven under certain physically reasonable conditions on the considered potential for scattering.
vii
Introduction
Our path to Scattering theory starts with the following problem:
Suppose we wish to determine properties of a person’s brain without intrusive testing. One possible way proceed would be to pass waves through the head, and observe the scattered waves which exit. We would like to know that this scattering information can return useful information, and so we will frame this as a mathematical problem. To find a starting point for such a theory, we may use the following thought experiment: Suppose n you stand at the edge of a large lake in R (this is mathematics, so the lake will of course be a ball). On top of this, suppose it is 2 a.m. and you are wearing your nicest shoes. In the middle of the lake, you can vaguely make out the shape of an object. Naturally you wish to determine properties of this object but you can’t see it very well due to the time of day, and you don’t want to ruin your shoes by swimming out to the object. All you can do is make splashes on the boundary of the lake, and measure the returning echoes (see Figure1). From this wave scattering information, would it be possible to gain any knowledge of the object?
Figure 1: You stand on the edge of a lake and splash at the mysterious object. The grey solid line is the incoming and reflected wave, and the grey dotted line is the path the wave would have taken if there was no object.
ix x
The mathematical framework behind trying to solve such a problem is known as scattering theory. There are a few varieties of scattering theory, and the two we’ll focus on are potential and obstacle scattering, which both fall under the heading of concrete scattering theory. Potential scattering can be thought of as modelling waves passing through localised inhomogeneity in a media (such as a change of refractive index), while obstacle scattering can be thought of as modelling waves travelling through space with some forbidden region (such as electromagnetic waves travelling through a region with a conductive chunk of metal in the way). Such a theory clearly has many real world applications, and it also has many interesting mathe- matical results! One such result is the spectral duality result of Eckmann and Pillet, given in their paper [EP95]. This spectral duality roughly states that an obstacle D appears “almost invisible” to some incoming radiation at frequencies approaching a given frequency k0, if and only if D can support standing waves of the same frequency k0. The aim of this thesis is to explain the result of Eckmann and Pillet and to attempt to generalise the result to potential scattering. To do this, we shall start by covering some basic analysis facts, using [RS80] and [Are15] as main recource. These will be used to construct the Laplacian - a key component of scattering theory, and the Dirichlet to Neumann operators - a key component of our attempts to generalise the result of [EP95]. The Laplacian is a differential operator which governs the mathematical understanding of many physical processes. To us, its significance will be in passing from physical waves travelling through a lake to mathematical waves travelling through wherever the reader believes mathematics to live. The Dirichlet to Neumann operators relate the boundary values of some function satisfying certain physical laws on a domain, to the flux of said function through the boundary. These operators will be used to come up with a “local problem” to match the potential scattering problem, in the same way the standing wave problem matches the scattering off D. After constructing these operators, we will explain how to pass from our physical problem of mystery objects in lakes and brain scans to a mathematical framework, known as scattering theory (following [Mel95]). This will leave us with enough knowledge to tackle the proof of Eckmann and Pillet’s, of which we will cover the main ideas. Finally, in the last two chapters we shall develop some possible extensions of the spectral duality result to potential scattering. One of these extensions seems quite promising, and a few results concerning it are proven. The promising extension is given in terms of an operator, which we have named the “transmission operator”. The transmission operator shows many similarities to the scattering matrix, a key component of scattering theory and the result of [EP95]. A proposed analogue of the main result of [EP95] related to the transmission operator is given in Section 5.3, as ConjectureB. A slightly modified form of this conjecture is proven for central potentials in Section 6.1, which is a new result of the author’s. Many preliminary results towards a proof of ConjectureB are also proven as new results. All unreferenced results of Chapter6 are the author’s work besides Lemma 6.10, which was joint work with Professor Andrew Hassell. Contents
Acknowledgementsv
Abstract vii
Introduction ix
Notation xiii
1 Background Analysis Concepts1 1.1 Unbounded Operators...... 1 1.2 Differentiability and Holomorphicity in Banach Spaces...... 5 1.3 The Spectral Theorem...... 6 1.4 Forms...... 10 1.5 Groups and Semigroups...... 11
2 The Laplacian and Dirichlet to Neumann Operators 15 2.1 The Laplacian...... 15 2.1.1 The Laplacian in Spherical Coordinates...... 17 2.2 The Spectral Theorem applied to ∆...... 19 2.3 Dirichlet to Neumann Operators...... 20 2.3.1 Dirichlet to Neumann Operator with respect to −∆...... 21 2.3.2 Dirichlet to Neumann Operator with respect to −∆ + A ...... 22
3 Scattering Theory 25 3.1 Trivial Scattering...... 25 3.2 Non-Trivial Scattering...... 27 3.3 Inverse Scattering Problem...... 30 3.3.1 Potential Scattering Inverse Problem...... 30 3.3.2 Obstacle Scattering Inverse Problems...... 31 3.4 Abstract Scattering Theory...... 32
4 Review of Spectral Duality for Planar Billiards 35 4.1 Spectral Duality - Easy Version...... 35 4.2 Difficulties with a Converse...... 36
xi xii CONTENTS
4.3 Spectral Duality - Hard Version...... 37 4.3.1 Potential Theory...... 38
4.3.2 Relating Ak to SD(k)...... 39 4.3.3 Proof of Theorem 4.5...... 41
5 Generalisations of Spectral Duality to Potential Scattering 45 5.1 Dirichlet to Neumann Boundary Condition Generalisation...... 45 5.2 Comparison of Scattering Matrices Generalisation...... 46 5.3 Cayley Transform Generalisation...... 47 5.3.1 Occurrence in Literature, and Notational Conventions...... 50
−1 6 Properties of CV (k)C0(k) 53 6.1 The Central Potential Case...... 56 6.2 Invariance Under Change of Domain...... 58 6.3 Spectral Properties...... 59 6.3.1 Accumulation of Eigenvalues...... 60 6.3.2 Flow of Eigenvalues...... 64 6.3.3 1 as Eigenvalue...... 74
Bibliography 79 Notation
H A separable Hilbert space over R or C.
ONB Orthonormal basis.
(·, ·) An inner product on a given Hilbert space. If multiple Hilbert spaces are being considered, the inner product will be distinguished by subscripts. We use the convention that the inner product is conjugate linear in the second variable: (af, g) = a(f, g).
(·, ·,..., ·) A tuple of elements from a given set.
V A a vector space over R or C.
n |x| The standard Euclidean norm of x ∈ R , or the norm of a number x ∈ C. The distinction will be clear from the context.
||v|| The norm of a vector V in a Banach space. If multiple norms are considered, they will be distinguished by subscripts.
Γ(T ) The graph of the linear operator T .
D(T ) The domain of the linear operator T .
σ(T ) The spectrum of the linear operator T .
ρ(T ) The resolvent set of the linear operator T .
L(X), L(X,Y ) The set of bounded linear transformtations from Banach space X to itself. The set of bounded linear transformations from Banach space X to Banach space Y .
χΩ The characteristic function of the set Ω. I.e. for a set X containing Ω, χΩ : X → R is the function which is identically 1 on Ω, and 0 elsewhere.
∆ The functional Laplacian, with sign convention based on Pn ∂ . i=1 ∂xi ∆ The Dirichlet Laplacian on U ⊂ n, with sign convention based on Pn ∂ . U R i=1 ∂xi
xiii xiv NOTATION
F The Fourier transform.
Λ,C A classical Dirichlet to Neumann operator and its Cayley transform, respec- tively.
Λ, C A semi-classical Dirichlet to Neumann operator and its Cayley transform, re- spectively.
SV (k),SD(k) The scattering matrix at frequency k for potential V or obstacle D with Dirich- let boundary conditions, respectively.
1 2 n H (D),H (D) The p = 2 Sobolev spaces on D ⊂ R .
1 ∞ 1 H0 (D) The closure of Cc (D) in H (D).
Df The vector of weak partial derivatives of a function f in some Sobolev space (or for f differentiable). Chapter 1
Background Analysis Concepts
This chapter will review some basics of the analysis techniques necessary throughout the rest of this thesis. These basics are more advanced than what is typically shown in an undergraduate degree in mathematics, but are still not the main component of this thesis so results will be given without proof. Most of the information below, and the relevant proofs, can be found in either [RS80] or [Are15]. Basic knowledge of Hilbert spaces and bounded operators will be assumed (roughly at an Analysis 3 level). However, the unfamiliar reader may find the earlier chapters of [RS80] a good reference for these concepts. Knowledge of Sobolev spaces will also be assumed. See [Eva10] for information. We will use the same notation as in [Eva10], which is repeated in Notation.
1.1 Unbounded Operators
There are many nice properties and definitions concerning bounded operators on Hilbert spaces, such as the notions of self-adjointness and compactness, and the spectral theorem for compact self-adjoint operators. However, in the real world bounded operators are incredibly limited. From a physics and PDE point of view, we would often like to know things about differential operators on function spaces. But differential operators are typically not bounded, as functions may have small norm, but vary wildly. This leads one to consider operators which are not required to be bounded. This section will discuss the main definitions and results for such operators. First recall
Definition 1.1. A bounded operator on a Hilbert space H is a linear map T : H → H, such that there exists a constant C > 0 with ||T x|| ≤ C||x|| for all x ∈ X. The set of bounded operators on H is written L(H).
This definition is too restrictive for our needs. We would still like some symmetry with respect to the inner product to be satisfied by the operators we consider. According to the Hellinger- Toeplitz Theorem (see Section III.5 of [RS80]), any operator T defined on all of H such that (T φ, ψ) = (φ, T ψ) for all φ, ψ ∈ H must necessarily be bounded. So we suppose that unbounded operators may not necessarily be defined on all of the Hilbert space we are working on, as is the case with differential operators on L2 spaces.
1 2 CHAPTER 1. BACKGROUND ANALYSIS CONCEPTS
Definition 1.2. An operator A on H is a linear map from the domain of A, a linear subspace D(A) ⊂ H, to H.
Most of the operators we encounter will have dense domain, so this will be assumed unless specified otherwise. Such a general definition does lead to some issues with analysis, in that we know nothing about how the operator interacts with limits. So we make another definition, in an attempt to control limits.
Definition 1.3. The graph of an operator T is the set
Γ(A) := {hφ, Aφi|φ ∈ D(A)} ⊂ H × H.
The product space H × H can be equipped with the inner product
(hψ1, φ1i, hψ2, φ2i) := (ψ1, ψ2) + (φ1, φ2) making H × H into a Hilbert space. A is called closed if Γ(A) is closed in H × H.
Note that the graph of an operator is a subspace of H × H, and we can always identify an operator with its graph. Depending on the domain which A is defined on, A may not be closed. However, if the domain is adjusted, it might be possible to define a closed operator extending A.
Definition 1.4. Let A, A0 be operators on H. If Γ(A) ⊂ Γ(A0), then we say A0 is an extension of A, and write A ⊂ A0. That is, D(A) ⊂ D(A0) and A and A0 agree on D(A).
Definition 1.5. An operator A is called closable if it has a closed extension. A closable operator A has a smallest closed extension, which is denoted A.
One way to generate closed extensions of an operator A is to take the closure of its graph, Γ(A). The issue with this is that the closure of the graph may no longer be the graph of an operator, that is it may contain points of the form h0, ψi for ψ 6= 0. However, if A is closable, then it is easy to check that Γ(A) = Γ(A).
Definition 1.6. Let A be a densely defined operator on H. Let D(A∗) be the set of φ ∈ H for which there is an η ∈ H with
(Aψ, φ) = (ψ, η) for all ψ ∈ D(A).
For each φ ∈ D(A∗), we define T ∗φ = η as above. T ∗ is called the adjoint of T .
It is clear that the adjoint as defined above is an operator, and that if A ⊂ B, then B∗ ⊂ A∗. Note that if A did not have dense domain, then the above definition would not be well-defined (if D(A) is not dense, D(A)⊥ is non-trivial and we could add any element of D(A)⊥ to η while preserving the defining equality as above). In the case that A is a bounded operator, this definition agrees with the usual definition of the adjoint constructed using the Riesz representation theorem. But in the case of unbounded operators, D(A∗) can be quite messy (even possibly just {0}!). If the adjoint A∗ does have dense domain, then we can define the double adjoint A∗∗ := (A∗)∗. In that case, we have the following helpful theorem: 1.1. UNBOUNDED OPERATORS 3
Theorem 1.7 ([RS80], Theorem VIII.1). Let A be a densely defined operator on H. Then:
1. A∗ is closed.
2. A is closable if and only if D(A∗) is dense, in which case A = A∗∗.
3. If A is closable, then (A)∗ = A∗.
This theorem allows us to generally consider closable operators instead of their closures when considering things such as self-adjointness, which is typically very helpful in calculations.
Definition 1.8. Let A be a closed operator on H. A complex number λ is in the resolvent set of A, denoted ρ(A), if λI − A is a bijection of D(A) onto H with bounded inverse. If λ ∈ ρ(A), −1 Rλ(A) = (λI − A) is called the resolvent of A at λ
Theorem 1.9 (First Resolvent Formula, [RS80], Theorem VI.5). Let A be an operator on H. Then for λ, µ ∈ ρ(A), we have
Rλ(A) − Rµ(A) = (µ − λ)Rµ(A)Rλ(A).
The verification of this fact is a simple computation. There are a few other significant properties of the resolvent, which will be covered in the next section on differentiability and holomorphicity of Banach space valued functions.
Definition 1.10. Let A be a closed operator on H. The spectrum of A, denoted σ(A), is the compliment of ρ(A) in C. The spectrum is further divided into the point spectrum and the residual spectrum. The point spectrum consists of all those λ for which there exists a non- zero φ ∈ D(A) such that Aφ = λφ, in which case λ is called an eigenvalue of A, with φ the corresponding eigenvector (sometimes called eigenfunction, eigenstate). The residual spectrum is everything in the spectrum that is not an eigenvalue.
We may sometimes refer to the spectrum of a closable, but not necessarily closed, operator, in which case we will always mean the spectrum of the closure. Now we can define what it means for an operator to be self-adjoint. Note that this is a bit more complicated than in the bounded case, as we may have issues with domains. For this reason, we have an intermediate notion of a symmetric operator
Definition 1.11. A densely defined operator A on H is called symmetric if A ⊂ A∗. In other words (Aψ, φ) = (ψ, Aφ) for all φ, ψ ∈ D(A).
Definition 1.12. A is called self adjoint if A = A∗.
Note that by Theorem 1.7, A∗ is closed. So if A is symmetric then A ⊂ A∗, so A∗ is a closed extension of A, so must also extend A = A∗∗. So for symmetric operators we have A ⊂ A∗∗ ⊂ A∗, for closed symmetric operators we have A = A∗∗ ⊂ A∗, and for self adjoint operators we have A = A∗∗ = A∗. Note also that in the case A is bounded, this definition is equivalent to the usual definition of self adjointness for bounded operators. 4 CHAPTER 1. BACKGROUND ANALYSIS CONCEPTS
It turns out that this definition of self adjointness for unbounded operators is the correct one as these domain equalities allow one to form functional calculi, which we will see is incredibly important to scattering theory. It should be noted that actually checking an operator is self adjoint directly from the definition could be quite difficult. However, there are many nice theorems which give simple conditions to check self-adjointness, and to construct self adjoint extensions of symmetric operators. While these theorems give a wonderful classification of self adjoint extensions, they are not relevant to the material of this thesis. The interested reader should check [RS75]. As in the case of bounded self adjoint operators, it is easy to check that the spectrum of a self adjoint operator lies completely on the real axis. This will be important in the next section, when we define a functional calculus on self adjoint operators. There are a slew of other definitions concerning operators, which give corresponding results on the spectrum. We say a self adjoint operator A is positive if (Ax, x) ≥ 0 for all x ∈ D(A), or 2 that it is bounded below if there exists an M ∈ R with (Ax, x) ≥ M||x|| . Note that a positive operator is bounded below with M = 0. We will use the notation A ≥ 0 to mean that A is a positive operator, and A ≥ B to mean A − B ≥ 0. Given A bounded below by M (or A ≥ MI), it is quite easy to show that A − λI is invertible for any λ < M, and hence the spectrum of A is bounded below by M. The proof of this is given as an exercise at the end of Chapter 8 of [RS80]. We will combine this knowledge with the following theorem and information about compact operators to obtain one of the most necessary theorems of this thesis, concerning self adjoint operators bounded below, with compact resolvent.
Theorem 1.13 (Hilbert-Schmidt Theorem, [RS80], Theorem VI.16). Let K be a compact self adjoint operator on H. Then there exists an ONB {φn} for H so that Kφn = λnφn for each n, and λn ∈ R. If H is infinite dimensional, then λn → 0 as n → ∞.
Recall also that the set of compact operators, denoted Com(H), form an ideal in L(H), that is Com(H) is a subspace of L(H), and for any S, T ∈ L(H) and K ∈ Com(H), SKT ∈ Com(H). It is also common knowledge (amongst analysts) that Com(H) is norm closed and closed under taking adjoints. These properties are proven in [RS80]. Note that the first resolvent formula (Theorem 1.9) combined with these facts shows that if an operator A has compact resolvent Rλ(A) for some λ ∈ ρ(A), then Rµ(A) is compact for all µ ∈ ρ(A). We can thus refer to operators with compact resolvent at some given point in their resolvent set as operators with compact resolvent. These results can be combined to give the following desired theorem, concerning self adjoint operators, bounded below with compact resolvent.
Theorem 1.14. Let A be a self adjoint operator on H bounded below with compact resolvent. Then there exists an ONB {φn} for H consisting of eigenvectors of A with real eigenvalues {λn}. If H is infinite dimensional, then λn → ∞ as n → ∞.
The proof of this is quite simple, so will be outlined. Since A is bounded below, RM (A) exists for some M ∈ R sufficiently low. Since A has compact resolvent, RM (A) is compact. Using self adjointness of A it can be checked that RM (A) is self adjoint (which requires M ∈ R). Theorem 1.2. DIFFERENTIABILITY AND HOLOMORPHICITY IN BANACH SPACES 5
1.13 is then applied to RM (A) to obtain an ONB {φn} for H consisting of eigenvectors of RM (A) with eigenvalues {µn}. It is then easy to check that each φn is an eigenvalue of A, with eigenvalue λ = M − 1 . The convergence of eigenvalues is then proven by the analogous statement in n µn Theorem 1.13 and using the fact that A is bounded below. As one last result we include the following theorem, attributed to Riesz and Schauder, con- cerning compact operators not necessarily assumed to be self adjoint. This will play a major role throughout the later chapters of this thesis. Theorem 1.15 (Riesz-Schauder Theorem, [RS80], Theorem VI.15). Let K be a compact operator on H. Then σ(K) is discrete, having no limit points except possibly at 0. Further, any non-zero λ ∈ σ(K) is an eigenvalue of finite degeneracy.
1.2 Differentiability and Holomorphicity in Banach Spaces
There are many nuances to coming up with a definition of “differentiable” or “holomorphic” for Banach space valued functions. Differentiating operator valued functions and applying theorems of complex analysis will feature heavily in later chapters. Rather than giving a complete description of the different differentiability conditions, we will now give a very brief overview and some examples. For the most part, everything works as it does for complex valued functions. For a Hilbert space valued function f : U ⊂ R → H, the derivative is defined in the obvious way: f(t + h) − f(t) f 0(t) = lim . h→0 h It is easy to check that this notion of derivative satisfies the usual properties of a derivative, such as linearity and the following nice relationship with the inner product:
(f, g)0 = (f, g0) + (f 0, g).
The only Banach-space valued functions we will encounter will be operator valued. Typically these will have unbounded derivative, so we use the strong definition of the derivative. We say an operator valued function A on U ⊂ R has derivative A˙(t) at t ∈ U, an unbounded and densely defined operator on H, if for all v ∈ D A˙(t) the following holds:
A(t + h)v − A(t)v lim = A˙(t)v. h→0 h 2 2 Example 1.16. For t ∈ [0, ∞), let A(t): R → L ([0, 1]) be the operator sending the pair 2 (a, b) ∈ R to the solution of the following boundary value problem: ( d2 2 (− dx2 − t )u(x) = 0 on (0, 1) u(0) = a, u(1) = b.
˙ 2 2 Then at t0 ∈ [0, ∞), A(t0): R → L ([0, 1]) sends the pair (a, b) to the solution of: ( d2 2 (− dx2 − t0)u ˙(x) = 2t0u(x) on (0, 1) u˙(0) = 0, u˙(1) = 0. Which is found by differentiating the equation governing A(t)(a, b) with respect to t. 6 CHAPTER 1. BACKGROUND ANALYSIS CONCEPTS
We will also be confronted with situations in which understanding holomorphicity will be re- quired. For a Hilbert space valued function f from some open set U ⊂ C to H, holomorphicity is defined exactly as one would expect. Namely, for each z ∈ U the following limit is required to exist: f(h + z) − f(z) lim . |h|→0 h Hilbert space valued holomorphic functions enjoy many of the properties of complex valued holo- morphic functions, such as local expressions as power series, infinite regularity and integral formulae. It is much more difficult to define a notion of holomorphic operator valued functions (also called a holomorphic family of operators). If we restrict to functions with image in the bounded operators, then we may say such functions are holomorphic if they admit local norm convergent power series representations.
Example 1.17. For an unbounded operator T on H, the resolvent Rλ(T ) is holomorphic. To see this, fix some λ0 ∈ ρ(T ). If we then consider λ close to λ0, we can find
∞ ! X n n Rλ0 (T ) I + (λ0 − λ) [Rλ0 (T )] n=1 −1 is a norm convergent power series for |λ − λ0| < ||Rλ0 (T )|| , and is the inverse of (λI − T ). Hence ρ(T ) is open, and Rλ(T ) is a holomorphic operator valued function. As a corollary, the spectrum of any operator is closed. This example was appropriated from Theorem VI.5 of [RS80].
Note that if A(z) is a holomorphic family of bounded operators in L(X,Y ) for Banach spaces X,Y , then for all x ∈ X, y∗ ∈ Y ∗, both A(z)x and y∗(A(z)x) are holomorphic functions (Banach space valued and complex valued respectively). In fact, it can be shown that a family B(z) of bounded operators from X to Y is holomorphic if and only if for all x ∈ X the function B(z)x is holomorphic, if and only if for all y∗ ∈ Y ∗ the function y∗(B(z)x) is holomorphic. This equivalence makes it much easier to check holomorphicity. As mentioned at the beginning of Section 1.1, we will need to consider unbounded operators. The previous definition of holomorphicity clearly fails for unbounded operator valued functions so we will have to come up with a new one. This is done in [Kat95], Chapter 7. However, the constructions in [Kat95] are complicated and space consuming (over 60 pages!) so will be left to the reader to digest. Fortunately, every unbounded family of operators we will need to consider can be linked to a bounded family of operators via the spectral theorem of the next section, and so we may just bootleg the results from the bounded case.
1.3 The Spectral Theorem
In this section, we will outline the main results known as the Spectral Theorem. There are many different faces of the spectral theorem, and each is very powerful and relevant to the aims of this thesis. The first form of the spectral theorem allows us to “take functions” of a self-adjoint operator A on H. In the case that H is finite dimensional, it is a common result that H has an orthonormal 1.3. THE SPECTRAL THEOREM 7 basis {φi} given by eigenvectors of A, such that each φi has real eigenvalue λi. Then given any function f : {λi} → C, we can define f(A) by setting f(A)φi = f(λi)φi and extending via linearity. We would like to apply this idea to the case that H is infinite dimensional, and the following theorem says that we can.
Theorem 1.18 (Spectral Theorem - Functional Calculus Form, [RS80], Theorem VIII.5). Let A be a self adjoint operator on H. Then there exists a unique map φˆ from B(R), the bounded Borel functions on R, to L(H) so that:
1. φˆ is an algebraic *-homomorphism, i.e.for all f, g ∈ B(R), λ ∈ C,
φˆ(fg) = φˆ(f)φˆ(g) φˆ(λf) = λφˆ(f) φˆ(1) = I φˆ(f) = φˆ(f)∗.
2. ||φˆ(f)|| ≤ ||f||∞.
3. Suppose fn is a sequence of bounded Borel functions on R with hn(x) → x for each x ∈ R and |hn(x)| ≤ |x| for all x and n. Then for all ψ ∈ D(A), φˆ(hn)ψ → Aψ.
4. If hn → h pointwise and ||hn||∞ is bounded, then φˆ(hn) → φˆ(h) strongly.
5. If Aψ = λψ, then φˆ(f)ψ = f(λ)ψ.
We will often write φˆ(f) as f(A) in analogy with the construction in the finite dimensional case, and refer to φˆ as a “functional calculus”. One of the most important applications of this form of the spectral theorem will be discussed in the later section on groups and semigroups. There is an important point which is missed in the above theorem given the level of generality we have approached with. In the construction of the functional calculus for bounded operators, one can start with defining f(A) for continuous functions f, defined only on σ(A). These are then used to extend the functional calculus to include all bounded Borel functions on R. But the functional calculus is still only “supported” on σ(A), in the sense that f(A) = f ◦ χσ(A)(A). This also holds for the unbounded self-adjoint operators A, although in that case this implicit dependence on only σ(A) is hidden even deeper within the proofs. With this in mind, we have the following incredibly important theorem, a generalisation of property (v) in the previous theorem:
Theorem 1.19 (Spectral Mapping Theorem, [RS80], Theorem VII.1(e)). Let A be a self adjoint operator on H, and f be a bounded real valued Borel function on σ(A). Then σ(f ◦ χσ(A)(A)) is equal to the essential range of f, i.e.
{λ ∈ R; ∀ > 0, µ(m ∈ M; |f(m) − λ| < ) > 0}.
In the case f is continuous on σ(A), then σ(f ◦ χσ(A)(A)) = f(σ(A)).
The second guise of the spectral theorem we will consider is incredibly important to scattering theory, and essentially gives rise to the Fourier transform. 8 CHAPTER 1. BACKGROUND ANALYSIS CONCEPTS
Theorem 1.20 (Spectral Theorem - Multiplication Operator Form, [RS80], Theorem VIII.4). Let A be a self adjoint operator on H. Then there exists a measure space hM, µi with µ a finite measure, 2 a unitary operator U : H → L (M, dµ) and a real valued measurable function f : M → R which is finite µ-a.e. such that:
1. ψ ∈ D(A) if and only if f(·)(Uψ)(·) ∈ L2(M, dµ).
2. If φ ∈ U(D(A)), then (UAU −1φ)(m) = f(m)φ(m).
Remark 1.21. In applications, we will typically be able to find M and µ explicitly. In some instances, it will be more convenient to drop the finite assumption on µ, as we shall see in Chapter 2. In the context of Theorem 1.20, the image of f and the spectrum of A can be directly related. Suppose for example that f is identically equal to λ ∈ R on a measurable set B ⊂ M with µ(B) 6= 0. Then χB(m) 6= 0 and (f(m) − λ) χB(m) = 0. We find: