SPECTRAL THEORY FOR ORDERED SPACES
A Thesis Submitted in Partial Fulfilment of the Requirements for the Degree of Master of Science
by G. PRIYANGA
to the School of Mathematical Sciences National Institute of Science Education and Research
Bhubaneswar 16-05-2017 DECLARATION
I hereby declare that I am the sole author of this thesis, submitted in par- tial fulfillment of the requirements for a postgraduate degree from the National
Institute of Science Education and Research (NISER), Bhubaneswar. I authorize NISER to lend this thesis to other institutions or individuals for the purpose of scholarly research.
Signature of the Student Date: 16th May, 2017
The thesis work reported in the thesis entitled Spectral Theory for Ordered
Spaces was carried out under my supervision, in the school of Mathematical Sciences at NISER, Bhubaneswar, India.
Signature of the thesis supervisor
School: Mathematical Sciences Date: 16th May, 2017
ii ACKNOWLEDGEMENTS
I would like to express my deepest gratitude to my thesis supervisor Dr. Anil Karn for his invaluable support, guidance and patience during the two year project period. I am particularly thankful to him for nurturing my interest in functional analysis and operator algebras. Without this project and his excellent courses, I might have never appreciated analysis so much.
I express my warm gratitude to my teachers at NISER and summer project supervisors, for teaching me all the mathematics that I used in this thesis work. Ahugethankstomyparentsfortheircontinuoussupportandencouragement.I am also indebted to my classmates who request the authorities and extend the deadline for report submission, each time!
iii ABSTRACT
This thesis presents an order theoretic study of spectral theory, developed by Alfsen and Shultz, extending the commutative spectral theorem to its non- commutative version. In this project, we build a spectral theory and functional cal- culus for order unit spaces and explore how this generalised spectral theory fits in the scheme of commutative case (function spaces) as well as the non-commutative case (Jordan Algebras). Amotivatingexampleforthistheorycomesfromthespectraltheoremfor monotone complete CR(X) spaces, which we study first. The goal is to extend the order theoretic ideas involved here, to a more general setting, namely to or- dered spaces. So, we begin with an order unit space A, which is in separating order and norm duality with a base norm space V. Here, we define maps called compressions, which give rise to projective units in A. Using projective units and projective faces, we develop various notions like comparability, orthogonality, com- patibility and also obtain certain lattice structures, under an assumption called standing hypothesis.Then,weconstructanabstractnotionofrangeprojection in A, which leads to a spectral decomposition result and functional calculus, for spaces satisfying a spectral duality condition. In the last section, we focus onto the spectral theorem for JBW-algebras and understand how the spectral theory works in non-commutative framework. Along the way, we also see concrete example of the order theoretic constructions developed above.
iv Contents
Introduction 1
1 Spectral Theory for Function Spaces 5
1.1 Range Projections ...... 7 1.2 Spectral Theorem ...... 11
2 Order Structure 15
2.1 Ordered Vector Spaces ...... 15 2.1.1 OrderUnitSpaces ...... 17 2.1.2 BaseNormSpaces ...... 22
2.2 Duality between Order Unit Space and Base Norm Space ...... 24
3 Spectral Theory for Ordered Spaces 30 3.1 Projections ...... 31
3.1.1 TangentSpacesandSemi-exposedFaces ...... 31 3.1.2 Smooth Projections ...... 35 3.2 Compressions ...... 41
3.2.1 ProjectiveUnitsandProjectiveFaces ...... 47 3.3 Relation between Compressions ...... 54
3.3.1 Comparability ...... 54 3.3.2 Orthogonality ...... 55 3.3.3 Compatibility ...... 57
v CONTENTS
3.4 The Lattice of Compressions ...... 65
3.4.1 The lattice of compressions when A = V ⇤ ...... 81 3.5 Range Projections ...... 87 3.6 SpacesinSpectralDuality ...... 91
3.7 Spectral Theorem ...... 107 3.7.1 Functional Calculus ...... 111
4 Spectral Theory for Jordan Algebras 114
4.1 OrderUnitAlgebras ...... 115 4.1.1 CharacterisingOrderUnitAlgebras ...... 122 4.1.2 Spectral Result ...... 127
4.2 JordanAlgberas...... 129 4.2.1 The Continuous Functional Calculus ...... 136 4.2.2 Triple Product ...... 140
4.2.3 Projections and Compressions in JB-algebras...... 145 4.2.4 Orthogonality ...... 150 4.2.5 Commutativity ...... 151
4.3 JBWalgebras ...... 158 4.3.1 Range Projections ...... 166 4.3.2 Spectral Resolutions ...... 172
5 Appendix 176
vi Introduction
The theory of von-Neumann algebras can be seen as a non-commutative generali- sation of integration theory. More precisely, as the lebesgue integral of a function is approximated by a sequence of simple functions, every self-adjoint element of a von-Neumann algebra can be written as a limit of finite linear sum over orthogonal projections. This idea is known as Spectral Decomposition Theory.Theobjective of this thesis work is to understand the order theoretic aspect of spectral theory, in the commutative case as well as non-commutative case.
The generalised spectral theory for ordered spaces, presented in this thesis, was originally developed by Erik M. Alfsen and Frederic W. Shultz, in the late 20th century. Motivated by structures in quantum mechanics, Alfsen and Shultz were interested in characterising the state space of operator algebras, particularly
Jordan algebras and C⇤-algebras. As part of this larger project, they established aspectraltheoryandfunctionalcalculusfororderunitspaces,whichgeneralised the corresponding results for von-Neumann algebras and JBW-algebras. This generalised theory seems to have applications in quantum mechanics. For instance, in the standard algebra model of quantum mechanical measurement, observables are represented by self-adjoint elements of a von Neumann algebra M and the states are represented by the elements in the normal state space K of M. Under the new generalisation, the basic concepts of this theory can be studied in a broader order-theoretic context: by representing states as elements in the distinguished base K of a base norm space V and observables by elements in the
1 CONTENTS
order unit space A = V ⇤. However, in my thesis work, the interest is not in state space characterisation or investigating its applications in physics. The focus of this project is developing ageneralspectraltheoryforsuitableorderedspacesandunderstandinghowthis extends the commutative spectral theorem to its non-commutative version. In this project, we mainly work with an order unit space (A, A+, e)whichisin separating order and norm duality with a base norm space (V,V+,K).Oninves- tigating the spectral theorem for monotone complete CR(X) spaces, which is one of the primary motivating example for this theory, one learns that projections and their orthogonality is the fundamental object involved in developing the spectral result. In order to construct these notions in a more general setting, we define maps, called compressions, on the order unit space A. The image of the order unit e under these compressions (known as projective unit) behave like projections in
A. We then explore various properties of the projective units and projective faces and develop notions of comparability, orthogonality and compatibility between compressions.
We further find that under an assumption called standing hypothesis,thesetof compressions on A form a lattice. Using the properties of this lattice structure, we then construct order theoretic tools, called range projections,analogoustothose in CR(X) spaces. Finally, we specialise to spaces in spectral duality,wherethese range projections lead to a spectral decomposition theory and functional calculus on A.
So, we obtain an abstract spectral theorem for order unit spaces. But how do these order theoretic constructions look in particular cases? A commutative example of this theory is seen in the case of monotone complete CR(X) spaces. An- other concrete example of this theory appears in the non-commutative framework of Jordan algebras. So, the last part of this thesis is devoted to understanding
2 CONTENTS the spectral theory for JBW-algebras. Here, we study about JB-algebras through order unit algebras, see examples of concrete compression (Up)andlearnhowthe spectral theory works in a non-commutative setting. In summary, we develop an order theoretic model of spectral theory for order unit spaces and explore how this generalised spectral theory fits in the scheme of commutative case (function spaces) as well as the non-commutative case (Jordan Algebras).
We now briefly describe the contents of each chapter.
Chapter 1 presents the order theoretic study of spectral theorem for monotone complete CR(X) spaces (continuous functions on a compact Hausdor↵space X). The spectral result and range projections constructed in this chapter are again used in Chapter 4, while developing the spectral theory for JBW-algebras. This material is mostly based on Alfsen and Shultz’s first book [2] (chapter 1).
Chapter 2 introduces order structure on vector spaces. Here, we explore vector order, cones, order unit space and base norm space. This is followed by a discussion on duality: dual pair, order duality, norm duality and then we show that order unit spaces and base norm spaces are dual to each other.
Chapter 3 focusses on the construction and development of the main spectral theorem. It begins with definitions and results on tangent spaces, semi-exposed faces and smooth projections in cones. This is followed by a discussion on general compressions, associated projective units, projective faces and relations between compressions: comparability, orthogonality and compatibility. Then, we study lattice structure on compressions, construction of abstract range projections and characterisation of spectral duality. We conclude by presenting the generalised spectral decomposition result and functional calculus for order unit spaces. This
3 CONTENTS content is largely based on Alfsen and Shultz’s second book [3] (chapters 7,8).
The last chapter is devoted to the spectral theory for JBW-algebras. Starting from the theory of order unit algebras, we obtain a continuous functional calculus on JB-algebras. Then, we look at triple products, construct concrete compressions and develop notions of orthogonality and commutativity in JB-algebra . Finally, we enter into JBW-algebras: basic definitions and relevant topologies. Here, we con- struct abstract range projections and obtain a spectral theorem for JBW-algebras, derived from the spectral theorem for monotone complete CR(X) . The topics on JB-algebra and JBW-algbera are presented from chapters 1, 2 of [3], while the section on order unit algebra is based on chapter 1 of [2].
Finally, a caveat: any errors found in this thesis are entirely my own!
4 Chapter 1
Spectral Theory for Function Spaces
We begin with the study of a spectral theory which is valid for a class of spaces of the form CR(X). The spectral theory for function spaces gives an example of spectral decomposition in the commutative case and serves as a motivation for the general theory. So, this chapter is devoted to an order theoretic understanding of the spectral theorem for monotone complete CR(X) spaces. The vector lattice
CR(X) is not monotone complete in general. However, certain important repre- sentation theorems for abstract algebras (such as the commutative von Neumann algebras and the normed Jordan algebras known as JBW-algebras) give rise to compact Hausdor↵spaces X, for which CR(X) is monotone complete. In the following pages, we will show that every element of a monotone com- plete CR(X) space can be approximated in norm by a linear span of projections. Further, the family of projections associated with a given element is unique and is characterised by certain order properties (Theorem 1.13). The objective of the project is to use order theoretic ideas and generalise this spectral theorem to a larger class of order unit spaces.
Let X be a compact Hausdor↵space. Let CR(X) denote the set of all continuous
5 1 Spectral Theory for Function Spaces
real valued functions on X.DefineapartialorderonCR(X) as follows:
f g f(x) g(x), x X () 8 2
1 Then, CR(X) forms a lattice with the following lattice operations
1 1 f g = (f + g + f g ),f g = (f + g f g ), for f,g C (X) _ 2 | | ^ 2 | | 2 R
Here, f denotes the usual modulus function f : X R defined as: | | | | ! f(x)iff(x) 0 f (x)= | | 8 f(x)iff(x) < 0 < Notation 1. Let f C (X) and: let 0 denote the constant 0 function on X. 2 R + Define f = f 0andf = (f 0). _ ^
+ + Remark 1.1. If f C (X) , then f = f f , f = f + f 2 R | |
Definition 1.2. A lattice L is said to be monotone complete if every bounded increasing (decreasing) net has a least upper bound (greatest lower bound) in L.
Example 1.3. If X = 1, 2 ...100 with the discrete topology, then C (X)is { } R monotone complete.
Throughout this chapter, we will assume that X is a compact Hausdor↵space and CR(X) is monotone complete. Infact, CR(X) is monotone complete precisely when X is extremally disconnected (i.e. the closure of every open set is open). One of the motivation to study such spaces, comes from the theory of von-Neumann algebra. For example, if A is a von-Neumann algebra and a A, then the spectrum 2 of a,denotedby (a), is extremally disconnected. Hence, the space CR( (a)) is monotone complete.
Remark 1.4. Every von-Neumann algebra is monotone complete.
1If f C (X) , then f also belongs to C (X) 2 R | | R
6 1 Spectral Theory for Function Spaces
Our goal is to develop a spectral result for monotone complete CR(X) spaces. We begin with some notations and definitions.
+ Notation 2. C (X) = f C (X) f 0 R { 2 R | } Notation 3. Let E X.Then E : X R is defined as ✓ ! 1ifx E (x)= 2 E 8 < 0otherwise Notation 4. Let f C (X) . The face: generated by f + in C (X)+, denoted by 2 R R face (f +), is the set g C (X) 0 g f +, for some 0 . { 2 R | } 1.1 Range Projections
Our first proposition proves the existence of range projections (defined later).
Proposition 1.5. Let X be a compact Hausdor↵space and assume that CR(X) is monotone complete. Then for each a C (X), the set E = s X a(s) > 0 is 2 R { 2 | } both closed and open in X. Further,
1. a 0 on E and a 0 on X r E. Also, E is the smallest closed subset of X such that a 0 on X r E.
+ + 2. E is the smallest closed subset of X for which Ea = a (pointwise prod- uct).
+ 3. E is the supremum in CR(X) of an increasing sequence in face (a ).
Proof. Let Y = s X a(s) > 0 .ThenY = E.Forn =1, 2, 3,... define { 2 | } fn : R [0, 1] as ! 0,x0 1 8 nx, 0 x fn(x)=> > n <> 1 1,x n > > :> 7 1 Spectral Theory for Function Spaces
Then, f a 1 is an increasing sequence in C (X), bounded above by the { n }n=1 R constant function 1. As C (X)ismonotonecomplete, f a b,forsome R { n } ! b C (X). 2 R
Claim. b = E
1 s Y = n N such that a(s) .Thisimplies(fn a)(s)=1and 2 )9 2 n hence, b(s)=1.Thisistrueforalls Y .So,b(Y ) = 1. And as b is continuous, 2 we have
b(E)=b(Y )=1
Next, fix t X r E.Choose2 c C (X)suchthatc(X) [0, 1],c(t)= 2 2 R ✓ 0,c(E)=1. If s E,then(f a)(s) 1=c(s), n.Thus,b(s) c(s), s E. 2 n 8 8 2 If s/E,then(f a)(s)=0, n.Thus,0=b(s) c(s), s/E. 2 n 8 8 2 Hence, b c.Inparticular,0 b(t) c(t)=0.So,b(t)=0.Sincet X r E 2 was arbitrary, we have
b(X r E)=0
Hence, b = E.
1 c 1 As, b is continuous, E = b (1) and E = b (0) are closed subsets of X. Hence, E is both open and closed in X.
1. As a>0onY ,wehavea 0onE = Y .Ifs/E,thens/Y = a(s) 0. 2 2 ) Thus, a 0onX r E. Next, let F be a closed subset of X such that a 0onX r F . Now, s Y = a(s) > 0= s/X r F = s F . Hence, Y F = 2 ) ) 2 ) 2 ✓ ) Y = E F .Thus,E is the smallest closed subset of X such that a 0on ✓ X r E. 2Compact Hausdor↵spaces are normal. Hence, it is possible to separate a point and a closed set by a continuous function.
8 1 Spectral Theory for Function Spaces
+ + 2. We will prove that Ea = a . For s E, (s)a+(s)=1.a+(s)=a+(s). 2 E For s/E, a(s) 0= a+(s)=0.Thus, (s)a+(s)=0.0=0=a+(s). 2 ) E Hence, a+(s)=a+(s), s X. E 8 2 + + Next, assume F is a closed subset of X such that F a = a .Then,for s Y , a(s)=a+(s) > 0. This implies (s)=1 = s F . Hence, 2 F ) 2 Y F = Y = E F . ✓ ) ✓
3. Note that a(s),sE a+(s)= 2 8 < 0, otherwise and for the f defined previously, { n} :
0,a(s) 0(i.e.s/E) 2 1 8 na(s), 0 a(s) (fn a)(s)=fn(a(s)) = > > n <> 1 1,a(s) n > Thus, f a C (X)+ and (f a>)(s) na+(s), s X. Hence, f a n 2 R n : 8 2 n na+ = f a face (a+). Further, we have shown that f a ) n 2 { n }%
b = E.Therefore, E is the supremum in CR(X)oftheincreasingsequence
+ f a 1 in face (a ). { n }n=1
2 Definition 1.6. An element p of CR(X)iscalledaprojectionifp = p .
Notation 5. For a pro jection p in C (X), denote p0 =1 p,where1isthe R function which takes the constant value 1 on X. Note that p0 is also a projection in CR(X).
Remark 1.7. Note that if p is a projection C (X), then p = for some closed 2 R E and open subset E of X.
9 1 Spectral Theory for Function Spaces
This is because if s X,thenp(s)=p(s).p(s)= p(s)=0orp(s)=1. 2 ) Therefore, p(X)= 0, 1 .DefineE = s X p(s)=1 .Thenp = . Also, { } { 2 | } E 1 1 as p is a continuous function on X,thesetsE = p (1) and X r E = p (0) are closed subsets of X. Hence, E is both open and closed in X.
Definition 1.8. Let a C (X)+.Definer(a)tobetheleastprojectionp in 2 R
CR(X)suchthatpa = a.Wecallr(a)tobetherange projection of a.
Proposition 1.5 implies the existence of range projections for positive elements of C (X)(a 0 a+ = a). Infact, for a C (X)+,wehaver(a)= R () 2 R E where E = s X a(s) > 0 . { 2 | }
Remark 1.9. If p, q are two projections in CR(X)oftheform E, F respectively (for some clopen subsets E, F of X), then p q E F . () ✓
Lemma 1.10. Let X be a compact Hausdor↵space and assume that CR(X) is monotone complete. If p is an increasing net of projections in C (X) and { ↵} R p p C (X), then p is also a projection in C (X). Similarly, if p is a ↵ % 2 R R { ↵} decreasing net of projections in C (X) and p p C (X), then p is a projection. R ↵ & 2 R
Proof. First consider p p. Note that 0 p 1, ↵ = 0 p 1. { ↵}% ↵ 8 ) Thus, p2 p. Next, 0 p p = p2 p2.Butp2 = p , ↵.Thus, ↵ ) ↵ ↵ ↵ 8 p =sup p =sup p2 p2.Therefore,p2 = p. ↵ ↵ ↵ ↵ Now, let p p.Then 1 p (1 p). Then, as above, (1 p)isa { ↵}& { ↵}% projection and hence p is also a projection.
Lemma 1.11. Let X be a compact Hausdor↵space and assume that CR(X) is monotone complete. Let a C (X) and p be a net of projections in C (X).If 2 R { ↵} R p p and p a 0, ↵, then pa 0. ↵ % ↵ 8
+ + Proof. Consider a = a a and p a =(p a) (p a) . ↵ ↵ ↵
Claim. p↵a =(p↵a)
10 1 Spectral Theory for Function Spaces
Let s X. As p is positive and takes values only 0 or 1, we have p (s)a (s)= 2 ↵ ↵ p (s)max a(s), 0 =max a(s)p (s), 0.p (s) =max a(s)p (s), 0 =(p a) (s). ↵ { } { ↵ ↵ } { ↵ } ↵ Hence, the claim.
+ Now as p a 0, we get p a =(p a) and (p a) = 0. Note that p a ↵ ↵ ↵ ↵ ↵ ! pa in C (X), because a is positive and bounded. But p a =0, ↵. Hence, R ↵ 8 + + + pa =0.So,pa = p(a a )=pa pa = pa 0. Thus, pa 0.
Lemma 1.12. Let X be a compact Hausdor↵space such that CR(X) is monotone complete. Let a C (X) and R.Ifp is a projection in C (X) such that 2 R 2 R pa p, then p 1 r (a 1)+ . Proof. Note that pa p = p(a 1) 0. Let b = a 1. Then pb 0= ) ) (pb)+ =0.But(pb)+ = pb+ (as proved in the claim of the previous lemma). Thus,
+ + + + pb =0 = p0b = b . Hence, r(b ) p0 (by proposition 1.5 (2)). Therefore, ) p 1 r (a 1)+ . 1.2 Spectral Theorem
We are now ready to prove our main spectral result for monotone complete CR(X) spaces.
Theorem 1.13. Let X be a compact Hausdor↵space such that CR(X) is monotone complete and let a CR(X). Then, there is a unique family e R of projections 2 { } 2 in CR(X) such that
(i) e a e ,e0 a e0 , R 8 2
(ii) e =0for < a ,e=1for > a k k k k
(iii) e e for <µ µ
(iv) eµ = e , R µ> 2 V 11 1 Spectral Theory for Function Spaces
The family e is given by e =1 r (a 1)+ . { } Further, for each increasing finite sequence = , ... with < a { 0 1 n} 0 k k n and n > a , define = max1 i n( i i 1) and s = i=1 i(e i e i 1 ). k k k k Then, P
lim s a =0 0 k k k k!
+ Proof. Define e =1 r (a 1) and E = s X a(s) > .Thene0 = { 2 | } + 1 e = r (a 1) = E and e =1 E = Ec . (i) By proposition 1.5, we know that
a 1 0onE ,a 1 0onEc (1.2.1) c e =0onE ,e =1onE (1.2.2)
Using the above and the fact that e is positive, we get that e (a 1) 0 on X.Thus, e a e , 8 Similarly,
c c e 0 =0onE ,e 0 =1onE (1.2.3)
Combining 1.2.3 with 1.2.1, we get
e0 a e0 , 8
(ii) When > a ,wehaveE = s X a(s) > > a = = e = k k { 2 | k k} ; ) c =1. E When < a ,wehaveE = s X a(s) a > = X = e = k k { 2 | k k } ) c =0. E
(iii) <µ = s X a(s) > s X a(s) >µ = E E = ){ 2 | }◆{ 2 | } ) ◆ µ ) c c E E = Ec Ec = e eµ. ✓ µ ) µ )
12 1 Spectral Theory for Function Spaces
(iv) Fix R.From(ii),(iii),weknowthateµ =1, µ> a and eµ 2 8 k k
e , µ> .Takeµ0 > a .Then eµ µ0 µ> is a decreasing net of pro- 8 k k { }
jections in CR(X), bounded below by e .Bymonotonecompletenessand
lemma 1.10, eµ µ0 µ0> p ,forsomeprojectionp in CR(X). Infact, { } &
p = µ> eµ and thus V p e Let E = s X p(s)=0 . Note that p(s)=1 = e (s)=1, µ> . { 2 | } ) µ 8 Hence, Ec Ec , µ> .Thus,by(1.2.1),a µ1onEc, µ> .This ✓ µ 8 8 implies a 1onEc.Thus,pa p ( p =1onEc). Now by lemma 1.12, we get p e
So, p = e .
Uniqueness:
Let f R be another family of projections in CR(X)satisfying(i),(ii), { } 2 (iii) and (iv). In particular, f a f , .Thus,bylemma1.12,weget 8
f e , (1.2.4) 8
For each R,defineE = s X e (s)=0 and F = s X f (s)= 2 { 2 | } { 2 | 0 .By(1.2.4),wegetE F . Also, by (i), } ✓
a 1onE ,a 1onEc a µ1onF ,a µ1onF c µ µ
For <µand x F ,wehavea(x) µ> = x E .Thus, 2 µ ) 2
Fµ E , µ> .Bute = Ec and fµ = F c .So,byremark1.9,weget ✓ 8 µ e f , µ> .Then,e f = f .Thus,e f , . µ 8 µ> µ 8 V Hence, f = e , R. 8 2
13 1 Spectral Theory for Function Spaces
Now, let = , ... be a finite increasing sequence of real numbers { 0 1 n} with < a and > a .DefineE = s X e (s)=1 ,i=1, 2 ...n. 0 k k n k k i { 2 | i } Then by (ii) and (iii), we see that = E E E ... E = X.Define ; 0 ✓ 1 ✓ 2 ✓ ✓ n n s = i=1 i(e i e i 1 ). Then x Ei r Ei 1 = s (x)= i. Also, as a 2 ) P c c i1onE i and a i 11onE i 1 ,wehave i 1 a i on Ei Ei 1 = EirEi 1. \ So,we get 0 s a i i 1 on Ei r Ei 1 which implies s (x) a(x) k k