Spectral Properties of an Operator-Fractal Keri Kornelson University of Oklahoma - Norman NSA Texas A&M University July 18, 2012 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 1 / 25 Acknowledgements This is joint work with Palle Jorgensen (University of Iowa) Karen Shuman (Grinnell College). K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 2 / 25 Outline 1 Bernoulli-Cantor Measures 2 Fourier bases 3 Families of ONBs 4 Operator-fractal K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 3 / 25 Bernoulli-Cantor Measures Iterated Function Systems (IFSs) We will construct an L2 space via an iterated function system. Definition An Iterated Function System (IFS) is a finite collection {τi }ki=1 of contractive maps on a complete metric space. The map on the compact subsets given by A 7→ k [ τi (A) i=1 is a contraction in the Hausdorff metric. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 4 / 25 Bernoulli-Cantor Measures IFS Attractor Set By the Banach Contraction Mapping Theorem, there exists a unique “fixed point” of the map. In other words, there is a compact set X satisfying the invariance relation: k [ τi (X ) = X . (1) i=1 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 5 / 25 Bernoulli-Cantor Measures IFS Attractor Set By the Banach Contraction Mapping Theorem, there exists a unique “fixed point” of the map. In other words, there is a compact set X satisfying the invariance relation: k [ τi (X ) = X . (1) i=1 The set X is called the attractor of the IFS. We say (1) is an invariance held by X . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 5 / 25 Bernoulli-Cantor Measures IFS Attractor Set By the Banach Contraction Mapping Theorem, there exists a unique “fixed point” of the map. In other words, there is a compact set X satisfying the invariance relation: k [ τi (X ) = X . (1) i=1 The set X is called the attractor of the IFS. We say (1) is an invariance held by X . Given any compact set A0 , successive iterations of our contraction An+1 = k [ τi (An ) i=1 converge (in Hausdorff metric) to the attractor X . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 5 / 25 Bernoulli-Cantor Measures Examples The Cantor ternary set in R: τ0 (x ) = K. Kornelson (U. Oklahoma) 1 x 3 τ1 (x ) = Operator-Fractal 1 2 x+ 3 3 TAMU 07/18/2012 6 / 25 Bernoulli-Cantor Measures Examples The Cantor ternary set in R: τ0 (x ) = 1 x 3 τ1 (x ) = 1 2 x+ 3 3 The Sierpinski gasket in R2 . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 6 / 25 Bernoulli-Cantor Measures Examples The Cantor ternary set in R: τ0 (x ) = 1 x 3 τ1 (x ) = 1 2 x+ 3 3 The Sierpinski gasket in R2 . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 6 / 25 Bernoulli-Cantor Measures Examples The Cantor ternary set in R: τ0 (x ) = 1 x 3 τ1 (x ) = 1 2 x+ 3 3 The Sierpinski gasket in R2 . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 6 / 25 Bernoulli-Cantor Measures Examples The Cantor ternary set in R: τ0 (x ) = 1 x 3 τ1 (x ) = 1 2 x+ 3 3 The Sierpinski gasket in R2 . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 6 / 25 Bernoulli-Cantor Measures Examples The Cantor ternary set in R: τ0 (x ) = 1 x 3 τ1 (x ) = 1 2 x+ 3 3 The Sierpinski gasket in R2 . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 6 / 25 Bernoulli-Cantor Measures IFS Measure Hutchinson, 1981 {τi }ki=1 an IFS {pi }ki=1 probability weights, i.e. pi ≥ 0, Define a map on measures: ν 7→ k X Pk i=1 pi = 1 pi (ν ◦ τi−1 ) (2) i=1 The Banach Theorem yields again a unique “fixed point”, in this case a probability measure supported on X . This measure µ is called an equilibrium or IFS measure for the IFS. As a fixed point, µ satisfies the invariance property: µ= k X pi (µ ◦ τi−1 ). (3) i=1 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 7 / 25 Bernoulli-Cantor Measures Bernoulli IFS Definition A Bernoulli IFS consists of two affine maps τ+ and τ− on R of the form τ+ (x ) = λ(x + 1) τ− (x ) = λ(x − 1) for 0 < λ < 1. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 8 / 25 Bernoulli-Cantor Measures Bernoulli IFS Definition A Bernoulli IFS consists of two affine maps τ+ and τ− on R of the form τ+ (x ) = λ(x + 1) τ− (x ) = λ(x − 1) for 0 < λ < 1. Bernoulli attractor set — Xλ K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 8 / 25 Bernoulli-Cantor Measures Bernoulli IFS Definition A Bernoulli IFS consists of two affine maps τ+ and τ− on R of the form τ+ (x ) = λ(x + 1) τ− (x ) = λ(x − 1) for 0 < λ < 1. Bernoulli attractor set — Xλ Bernoulli convolution measure (p+ = p− = 12 ) — µλ — supported on Xλ . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 8 / 25 Bernoulli-Cantor Measures Bernoulli IFS Definition A Bernoulli IFS consists of two affine maps τ+ and τ− on R of the form τ+ (x ) = λ(x + 1) τ− (x ) = λ(x − 1) for 0 < λ < 1. Bernoulli attractor set — Xλ Bernoulli convolution measure (p+ = p− = 12 ) — µλ — supported on Xλ . Historical note: The Bernoulli measures date back to work of Erdös and others, long before this IFS approach came along. µλ is the distribution of the random variable P k k ±λ where + and − have equal probability. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 8 / 25 Fourier bases Fourier bases for Bernoulli measures? Consider the Hilbert space L2 (µλ ). K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 9 / 25 Fourier bases Fourier bases for Bernoulli measures? Consider the Hilbert space L2 (µλ ). Is it ever possible for L2 (µλ ) to have a Fourier basis, i.e. an orthonormal basis (ONB) of complex exponential functions? K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 9 / 25 Fourier bases Fourier bases for Bernoulli measures? Consider the Hilbert space L2 (µλ ). Is it ever possible for L2 (µλ ) to have a Fourier basis, i.e. an orthonormal basis (ONB) of complex exponential functions? If so, does the existence of a Fourier basis depend on λ? K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 9 / 25 Fourier bases Fourier bases for Bernoulli measures? Consider the Hilbert space L2 (µλ ). Is it ever possible for L2 (µλ ) to have a Fourier basis, i.e. an orthonormal basis (ONB) of complex exponential functions? If so, does the existence of a Fourier basis depend on λ? Given a spectral measure, what are the possible spectra? K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 9 / 25 Fourier bases Fourier bases for Bernoulli measures? Consider the Hilbert space L2 (µλ ). Is it ever possible for L2 (µλ ) to have a Fourier basis, i.e. an orthonormal basis (ONB) of complex exponential functions? If so, does the existence of a Fourier basis depend on λ? Given a spectral measure, what are the possible spectra? Could a non-spectral measure have a frame of exponential functions? K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 9 / 25 Fourier bases Getting started: µ bλ Recall that, given λ ∈ (0, 1), the Bernoulli IFS is: τ+ (x ) = λ(x + 1) and τ− (x ) = λ(x − 1). The Bernoulli measure µλ satisfies the invariance 1 −1 (µλ ◦ τ+−1 + µλ ◦ τ− ). 2 Then the Fourier transform of µλ is: µλ = µ bλ (t) = = = = .. . = Z e2πixt dµλ (x ) Z Z 1 1 e2πi(λx+λ)t dµλ (x ) + e2πi(λx−λ)t dµλ (x ) 2 2 cos(2πλt)b µλ (λt) cos(2πλt) cos(2πλ2 t)b µλ (λ2 t) .. . ∞ Y cos(2πλk t) k =1 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 10 / 25 Fourier bases Orthogonality Condition Denote by eγ the exponential function e2πiγ· in L2 (µλ ). K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 11 / 25 Fourier bases Orthogonality Condition Denote by eγ the exponential function e2πiγ· in L2 (µλ ). heγ , eγ̃ iL2 = Z eγ−γ̃ dµλ = µ bλ (γ − γ̃) ∞ Y k = cos 2π λ (γ − γ̃) k =1 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 11 / 25 Fourier bases Orthogonality Condition Denote by eγ the exponential function e2πiγ· in L2 (µλ ). heγ , eγ̃ iL2 = Z eγ−γ̃ dµλ = µ bλ (γ − γ̃) ∞ Y k = cos 2π λ (γ − γ̃) k =1 Lemma The two exponentials eγ , eγ̃ are orthogonal if and only if one of the factors in the infinite product above is zero. This is equivalent to 1 −k λ (2m + 1) : k ∈ N, m ∈ Z =: Zλ . γ − γ̃ ∈ 4 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 11 / 25 Fourier bases Surprising first results Theorem (Jorgensen, Pedersen 1998) L2 (µ 1 ) has an ONB of exponential functions. 4 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 12 / 25 Fourier bases Surprising first results Theorem (Jorgensen, Pedersen 1998) L2 (µ 1 ) has an ONB of exponential functions. 4 Example E(Γ 1 ) is an ONB for L2 (µ 1 ), where 4 Γ1 = 4 4 p X aj 4j : aj ∈ {0, 1}, p finite j=0 K. Kornelson (U. Oklahoma) = {0, 1, 4, 5, 16, 17, 20, 21, 64, . . .}. Operator-Fractal TAMU 07/18/2012 12 / 25 Fourier bases Another surprise: λ = 1 3 Theorem (Jorgensen, Pedersen 1998) Not only is there no Fourier basis when λ = 31 , but orthogonal collections of exponential functions can have at most 2 elements. There is also a more general version in the [JP1998] paper: K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 13 / 25 Fourier bases Another surprise: λ = 1 3 Theorem (Jorgensen, Pedersen 1998) Not only is there no Fourier basis when λ = 31 , but orthogonal collections of exponential functions can have at most 2 elements. There is also a more general version in the [JP1998] paper: Theorem (Jorgensen, Pedersen 1998) Given λ = n1 , if n is even, there is an ONB of exponentials for L2 (µ 1 ) but 2n when n is odd, there can be only finitely many elements in any orthogonal collection of exponentials. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 13 / 25 Fourier bases More recent progress Theorem (Jorgensen, K, Shuman 2008; Hu, Lau 2008) For λ = ba , if b is odd, then any orthonormal collection of exponentials in L2 (µλ ) must be finite. If b is even, then there exists countable collections of orthonormal exponentials in L2 (µλ ). K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 14 / 25 Fourier bases More recent progress Theorem (Jorgensen, K, Shuman 2008; Hu, Lau 2008) For λ = ba , if b is odd, then any orthonormal collection of exponentials in L2 (µλ ) must be finite. If b is even, then there exists countable collections of orthonormal exponentials in L2 (µλ ). Theorem (Dutkay, Han, Jorgensen 2009) If λ > 12 , i.e. there is essential overlap, then L2 (µλ ) does not have an ONB (or even a frame) of exponential functions. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 14 / 25 Fourier bases “Canonical” ONBs [Jorgensen, Pedersen 1998] Definition Let λ = 1 2n and consider the set from Jorgensen & Pedersen p X n no Γ1 = , p finite . aj (2n)j : aj ∈ 0, 2n 2 j=0 We call Γ 1 the canonical spectrum and E(Γ 1 ) the canonical ONB for L2 (µ 1 ). 2n 2n 2n Note: We will justify the nomenclature by describing alternate bases for the same spaces. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 15 / 25 Families of ONBs Families of ONBs Jorgensen, K, Shuman 2011 JFAA 1 . We can construct alternate orthogonal families of exponentials Let λ = 2n from the canonical ONBs E(Γ 1 ). We then determine whether these alternate 2n sets are ONBs. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 16 / 25 Families of ONBs Families of ONBs Jorgensen, K, Shuman 2011 JFAA 1 . We can construct alternate orthogonal families of exponentials Let λ = 2n from the canonical ONBs E(Γ 1 ). We then determine whether these alternate 2n sets are ONBs. Theorem The set E(3Γ 1 ) is an (alternate) orthonormal basis for L2 (µ 1 ). 8 K. Kornelson (U. Oklahoma) 8 Operator-Fractal TAMU 07/18/2012 16 / 25 Families of ONBs Families of ONBs Jorgensen, K, Shuman 2011 JFAA 1 . We can construct alternate orthogonal families of exponentials Let λ = 2n from the canonical ONBs E(Γ 1 ). We then determine whether these alternate 2n sets are ONBs. Theorem The set E(3Γ 1 ) is an (alternate) orthonormal basis for L2 (µ 1 ). 8 8 Theorem If p < 2(2n−1) , π then E(pΓ 1 ) is an ONB for L2 (µ 1 ). K. Kornelson (U. Oklahoma) 2n 2n Operator-Fractal TAMU 07/18/2012 16 / 25 Families of ONBs Families of ONBs Jorgensen, K, Shuman 2011 JFAA 1 . We can construct alternate orthogonal families of exponentials Let λ = 2n from the canonical ONBs E(Γ 1 ). We then determine whether these alternate 2n sets are ONBs. Theorem The set E(3Γ 1 ) is an (alternate) orthonormal basis for L2 (µ 1 ). 8 8 Theorem If p < 2(2n−1) , π then E(pΓ 1 ) is an ONB for L2 (µ 1 ). 2n 2n Laba/Wang and Dutkay/Jorgensen have described many other values of p for which pΓ 1 is a spectrum, particularly in the 41 case. 2n K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 16 / 25 Operator-fractal Operators on L2 Dutkay, Jorgensen, 2009: When λ = 41 , both Γ 1 and 5Γ 1 are spectra for L2 (µ 1 ). 4 4 4 Γ 1 = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} 4 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 17 / 25 Operator-fractal Operators on L2 Dutkay, Jorgensen, 2009: When λ = 41 , both Γ 1 and 5Γ 1 are spectra for L2 (µ 1 ). 4 4 4 Γ 1 = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} 4 For each γ ∈ Γ 1 , define 4 S0 S1 U K. Kornelson (U. Oklahoma) : eγ 7→ e4γ : eγ → 7 e4γ+1 : eγ 7→ e5γ Operator-Fractal TAMU 07/18/2012 17 / 25 Operator-fractal Operators on L2 Dutkay, Jorgensen, 2009: When λ = 41 , both Γ 1 and 5Γ 1 are spectra for L2 (µ 1 ). 4 4 4 Γ 1 = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} 4 For each γ ∈ Γ 1 , define 4 S0 S1 U : eγ 7→ e4γ : eγ → 7 e4γ+1 : eγ 7→ e5γ S0 and S1 map between ONB elements, so are both isometries. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 17 / 25 Operator-fractal Operators on L2 Dutkay, Jorgensen, 2009: When λ = 41 , both Γ 1 and 5Γ 1 are spectra for L2 (µ 1 ). 4 4 4 Γ 1 = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} 4 For each γ ∈ Γ 1 , define 4 S0 S1 U : eγ 7→ e4γ : eγ → 7 e4γ+1 : eγ 7→ e5γ S0 and S1 map between ONB elements, so are both isometries. U maps one ONB to another, so U is a unitary operator. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 17 / 25 Operator-fractal Structure of Γ Example: Let λ = 1 4 and denote H = L2 (µ 1 ). 4 Γ = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} Γ = 4Γ ∪ (1 + 4Γ) = 42 Γ ∪ 4(1 + 4Γ) ∪ (1 + 4Γ) .. . = 4k Γ ∪ 4k −1 (1 + 4Γ) · · · 4(1 + 4Γ) ∪ (1 + 4Γ) S0 (H) is the span of the exponentials E(4Γ) K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 18 / 25 Operator-fractal Structure of Γ Example: Let λ = 1 4 and denote H = L2 (µ 1 ). 4 Γ = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} Γ = 4Γ ∪ (1 + 4Γ) = 42 Γ ∪ 4(1 + 4Γ) ∪ (1 + 4Γ) .. . = 4k Γ ∪ 4k −1 (1 + 4Γ) · · · 4(1 + 4Γ) ∪ (1 + 4Γ) S0 (H) is the span of the exponentials E(4Γ) S0k (H) is the span of E(4k Γ) K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 18 / 25 Operator-fractal Structure of Γ Example: Let λ = 1 4 and denote H = L2 (µ 1 ). 4 Γ = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} Γ = 4Γ ∪ (1 + 4Γ) = 42 Γ ∪ 4(1 + 4Γ) ∪ (1 + 4Γ) .. . = 4k Γ ∪ 4k −1 (1 + 4Γ) · · · 4(1 + 4Γ) ∪ (1 + 4Γ) S0 (H) is the span of the exponentials E(4Γ) S0k (H) is the span of E(4k Γ) H ⊃ S0 (H) ⊃ S02 (H) ⊃ · · · K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 18 / 25 Operator-fractal Structure of Γ Example: Let λ = 1 4 and denote H = L2 (µ 1 ). 4 Γ = {0, 1, 4, 5, 16, 17, 20, 21, 64, 65, . . .} Γ = 4Γ ∪ (1 + 4Γ) = 42 Γ ∪ 4(1 + 4Γ) ∪ (1 + 4Γ) .. . = 4k Γ ∪ 4k −1 (1 + 4Γ) · · · 4(1 + 4Γ) ∪ (1 + 4Γ) S0 (H) is the span of the exponentials E(4Γ) S0k (H) is the span of E(4k Γ) H ⊃ S0 (H) ⊃ S02 (H) ⊃ · · · If we define Wk = S0k (H) ⊖ S0k +1 (H), then H = sp(e0 ) ⊕ ∞ M Wk k =0 K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 18 / 25 Operator-fractal The operator U Recall U : eγ 7→ e5γ . How does that scaling by (×5) in the ONB frequencies interact with the inherent scaling (×4) of the fractal measure space? Theorem (Jorgensen,K,Shuman 2011) K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 19 / 25 Operator-fractal The operator U Recall U : eγ 7→ e5γ . How does that scaling by (×5) in the ONB frequencies interact with the inherent scaling (×4) of the fractal measure space? Theorem (Jorgensen,K,Shuman 2011) Each subspace Wk is invariant under U. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 19 / 25 Operator-fractal The operator U Recall U : eγ 7→ e5γ . How does that scaling by (×5) in the ONB frequencies interact with the inherent scaling (×4) of the fractal measure space? Theorem (Jorgensen,K,Shuman 2011) Each subspace Wk is invariant under U. With respect to the Wk ordering of Γ, the matrix of U has block diagonal form...and the infinite blocks are all the same! K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 19 / 25 Operator-fractal The operator U Recall U : eγ 7→ e5γ . How does that scaling by (×5) in the ONB frequencies interact with the inherent scaling (×4) of the fractal measure space? Theorem (Jorgensen,K,Shuman 2011) Each subspace Wk is invariant under U. With respect to the Wk ordering of Γ, the matrix of U has block diagonal form...and the infinite blocks are all the same! Even more, in the (×4, ×5) case U actually has a self-similar structure: U = (e0 ⊗ e0 ) ⊕ ∞ M Me1 U. k =1 We call U an operator-fractal. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 19 / 25 Operator-fractal Matrix of U 0 Γ0 Γ1 Γ2 Γ3 K. Kornelson (U. Oklahoma) 0 1 0 0 0 0 .. . Γ0 0 Me1 U 0 0 0 .. . Γ1 0 0 Me1 U 0 0 .. . Γ2 0 0 0 Me1 U 0 .. . Operator-Fractal Γ3 0 0 0 0 Me1 U .. . ··· ··· ··· ··· ··· ··· .. . TAMU 07/18/2012 20 / 25 Operator-fractal Spectral properties of U Jorgensen, K, Shuman 2012 Proposition (JKS 2012) U has the following properties. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 21 / 25 Operator-fractal Spectral properties of U Jorgensen, K, Shuman 2012 Proposition (JKS 2012) U has the following properties. 1 is the only eigenvalue of U. (Ue0 = e0 ) K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 21 / 25 Operator-fractal Spectral properties of U Jorgensen, K, Shuman 2012 Proposition (JKS 2012) U has the following properties. 1 is the only eigenvalue of U. (Ue0 = e0 ) U is not spatially implemented; i.e. is not of the form Uf = f ◦ τ for τ a point transformation on [0, 1]. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 21 / 25 Operator-fractal Spectral properties of U Jorgensen, K, Shuman 2012 Proposition (JKS 2012) U has the following properties. 1 is the only eigenvalue of U. (Ue0 = e0 ) U is not spatially implemented; i.e. is not of the form Uf = f ◦ τ for τ a point transformation on [0, 1]. Theorem (JKS 2012) U is an ergodic operator; i.e. if Uv = v then v = ce0 for some c ∈ C. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 21 / 25 Operator-fractal Nelson-like cyclic subspaces Assume that Uv = v for some kv k = 1, v ∈ / sp(e0 ). We modify the cyclic subspaces of Nelson for use with our unitary operator U. Given v 6= 0, H(v ) := sp{U k v : k ∈ Z} is the U-cyclic subspace for v . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 22 / 25 Operator-fractal Nelson-like cyclic subspaces Assume that Uv = v for some kv k = 1, v ∈ / sp(e0 ). We modify the cyclic subspaces of Nelson for use with our unitary operator U. Given v 6= 0, H(v ) := sp{U k v : k ∈ Z} is the U-cyclic subspace for v . Let E be the projection-valued measure for U and, given v , let mv be the real measure mv (A) = hE(A)v , v i. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 22 / 25 Operator-fractal Nelson-like cyclic subspaces Assume that Uv = v for some kv k = 1, v ∈ / sp(e0 ). We modify the cyclic subspaces of Nelson for use with our unitary operator U. Given v 6= 0, H(v ) := sp{U k v : k ∈ Z} is the U-cyclic subspace for v . Let E be the projection-valued measure for U and, given v , let mv be the real measure mv (A) = hE(A)v , v i. If kv k = 1, mv is a probability measure. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 22 / 25 Operator-fractal Nelson-like cyclic subspaces Assume that Uv = v for some kv k = 1, v ∈ / sp(e0 ). We modify the cyclic subspaces of Nelson for use with our unitary operator U. Given v 6= 0, H(v ) := sp{U k v : k ∈ Z} is the U-cyclic subspace for v . Let E be the projection-valued measure for U and, given v , let mv be the real measure mv (A) = hE(A)v , v i. If kv k = 1, mv is a probability measure. Theorem H(v ) = {φ(U)v : φ ∈ L2 (mv )}. Moreover, the map φ 7→ φ(U)v is an isometric isomorphism between L2 (mv ) and H(v ). K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 22 / 25 Operator-fractal Lemmas The following lemmas make heavy use of the isometric isomorphism between cyclic subspaces H(v ) and L2 (mv ). Lemma If Uv = v , then v is not in H(eγ ) for any γ ∈ Γ \ {0}. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 23 / 25 Operator-fractal Lemmas The following lemmas make heavy use of the isometric isomorphism between cyclic subspaces H(v ) and L2 (mv ). Lemma If Uv = v , then v is not in H(eγ ) for any γ ∈ Γ \ {0}. Lemma If v1 ⊥ H(v2 ), then H(v1 ) ⊥ H(v2 ). K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 23 / 25 Operator-fractal Lemmas The following lemmas make heavy use of the isometric isomorphism between cyclic subspaces H(v ) and L2 (mv ). Lemma If Uv = v , then v is not in H(eγ ) for any γ ∈ Γ \ {0}. Lemma If v1 ⊥ H(v2 ), then H(v1 ) ⊥ H(v2 ). Lemma If Uv = v for kv k = 1 then mv is a Dirac point mass at 1. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 23 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. There exists some γ ∈ Γ \ {0} such that hv , eγ i = 6 0. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. There exists some γ ∈ Γ \ {0} such that hv , eγ i = 6 0. Let v1 be the projection of v onto H(eγ ) and v2 = v − v1 . Note that v2 6= 0 and kv2 k2 = 1 − kv1 k2 . K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. There exists some γ ∈ Γ \ {0} such that hv , eγ i = 6 0. Let v1 be the projection of v onto H(eγ ) and v2 = v − v1 . Note that v2 6= 0 and kv2 k2 = 1 − kv1 k2 . H(v2 ) ⊥ H(eγ ) by one of the lemmas above. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. There exists some γ ∈ Γ \ {0} such that hv , eγ i = 6 0. Let v1 be the projection of v onto H(eγ ) and v2 = v − v1 . Note that v2 6= 0 and kv2 k2 = 1 − kv1 k2 . H(v2 ) ⊥ H(eγ ) by one of the lemmas above. mv (A) = hE(A)v1 , v1 i + hE(A)v2 , v2 i = mv1 (A) + mv2 (A). The cross terms are zero. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. There exists some γ ∈ Γ \ {0} such that hv , eγ i = 6 0. Let v1 be the projection of v onto H(eγ ) and v2 = v − v1 . Note that v2 6= 0 and kv2 k2 = 1 − kv1 k2 . H(v2 ) ⊥ H(eγ ) by one of the lemmas above. mv (A) = hE(A)v1 , v1 i + hE(A)v2 , v2 i = mv1 (A) + mv2 (A). The cross terms are zero. e v1 and m e v2 , giving Normalize to probability measures m e v1 (A) + kv2 k2 m e v2 (A). mv (A) = kv1 k2 m K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. There exists some γ ∈ Γ \ {0} such that hv , eγ i = 6 0. Let v1 be the projection of v onto H(eγ ) and v2 = v − v1 . Note that v2 6= 0 and kv2 k2 = 1 − kv1 k2 . H(v2 ) ⊥ H(eγ ) by one of the lemmas above. mv (A) = hE(A)v1 , v1 i + hE(A)v2 , v2 i = mv1 (A) + mv2 (A). The cross terms are zero. e v1 and m e v2 , giving Normalize to probability measures m e v1 (A) + kv2 k2 m e v2 (A). mv (A) = kv1 k2 m Thus mv is a convex combination of probability measures. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Operator-fractal Proof of ergodicity of U: Assume Uv = v for some v ∈ / sp(e0 ), kv k = 1. There exists some γ ∈ Γ \ {0} such that hv , eγ i = 6 0. Let v1 be the projection of v onto H(eγ ) and v2 = v − v1 . Note that v2 6= 0 and kv2 k2 = 1 − kv1 k2 . H(v2 ) ⊥ H(eγ ) by one of the lemmas above. mv (A) = hE(A)v1 , v1 i + hE(A)v2 , v2 i = mv1 (A) + mv2 (A). The cross terms are zero. e v1 and m e v2 , giving Normalize to probability measures m e v1 (A) + kv2 k2 m e v2 (A). mv (A) = kv1 k2 m Thus mv is a convex combination of probability measures. But, by hypothesis, mv is a Dirac point mass, hence is an extreme point in the space of probability measures. This gives a contradiction. K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 24 / 25 Thank You K. Kornelson (U. Oklahoma) Operator-Fractal TAMU 07/18/2012 25 / 25